spa.c revision 3e30c24aeefdee1631958ecf17f18da671781956
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
26 */
27
28/*
29 * SPA: Storage Pool Allocator
30 *
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 * pool.
34 */
35
36#include <sys/zfs_context.h>
37#include <sys/fm/fs/zfs.h>
38#include <sys/spa_impl.h>
39#include <sys/zio.h>
40#include <sys/zio_checksum.h>
41#include <sys/dmu.h>
42#include <sys/dmu_tx.h>
43#include <sys/zap.h>
44#include <sys/zil.h>
45#include <sys/ddt.h>
46#include <sys/vdev_impl.h>
47#include <sys/metaslab.h>
48#include <sys/metaslab_impl.h>
49#include <sys/uberblock_impl.h>
50#include <sys/txg.h>
51#include <sys/avl.h>
52#include <sys/dmu_traverse.h>
53#include <sys/dmu_objset.h>
54#include <sys/unique.h>
55#include <sys/dsl_pool.h>
56#include <sys/dsl_dataset.h>
57#include <sys/dsl_dir.h>
58#include <sys/dsl_prop.h>
59#include <sys/dsl_synctask.h>
60#include <sys/fs/zfs.h>
61#include <sys/arc.h>
62#include <sys/callb.h>
63#include <sys/systeminfo.h>
64#include <sys/spa_boot.h>
65#include <sys/zfs_ioctl.h>
66#include <sys/dsl_scan.h>
67#include <sys/zfeature.h>
68#include <sys/dsl_destroy.h>
69
70#ifdef	_KERNEL
71#include <sys/bootprops.h>
72#include <sys/callb.h>
73#include <sys/cpupart.h>
74#include <sys/pool.h>
75#include <sys/sysdc.h>
76#include <sys/zone.h>
77#endif	/* _KERNEL */
78
79#include "zfs_prop.h"
80#include "zfs_comutil.h"
81
82typedef enum zti_modes {
83	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
84	ZTI_MODE_ONLINE_PERCENT,	/* value is % of online CPUs */
85	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
86	ZTI_MODE_NULL,			/* don't create a taskq */
87	ZTI_NMODES
88} zti_modes_t;
89
90#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
91#define	ZTI_PCT(n)	{ ZTI_MODE_ONLINE_PERCENT, (n), 1 }
92#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
93#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
94
95#define	ZTI_N(n)	ZTI_P(n, 1)
96#define	ZTI_ONE		ZTI_N(1)
97
98typedef struct zio_taskq_info {
99	zti_modes_t zti_mode;
100	uint_t zti_value;
101	uint_t zti_count;
102} zio_taskq_info_t;
103
104static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
105	"issue", "issue_high", "intr", "intr_high"
106};
107
108/*
109 * This table defines the taskq settings for each ZFS I/O type. When
110 * initializing a pool, we use this table to create an appropriately sized
111 * taskq. Some operations are low volume and therefore have a small, static
112 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
113 * macros. Other operations process a large amount of data; the ZTI_BATCH
114 * macro causes us to create a taskq oriented for throughput. Some operations
115 * are so high frequency and short-lived that the taskq itself can become a a
116 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
117 * additional degree of parallelism specified by the number of threads per-
118 * taskq and the number of taskqs; when dispatching an event in this case, the
119 * particular taskq is chosen at random.
120 *
121 * The different taskq priorities are to handle the different contexts (issue
122 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
123 * need to be handled with minimum delay.
124 */
125const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
126	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
127	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
128	{ ZTI_N(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL }, /* READ */
129	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
130	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
131	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
132	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
133};
134
135static void spa_sync_version(void *arg, dmu_tx_t *tx);
136static void spa_sync_props(void *arg, dmu_tx_t *tx);
137static boolean_t spa_has_active_shared_spare(spa_t *spa);
138static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
139    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
140    char **ereport);
141static void spa_vdev_resilver_done(spa_t *spa);
142
143uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
144id_t		zio_taskq_psrset_bind = PS_NONE;
145boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
146uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
147
148boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
149extern int	zfs_sync_pass_deferred_free;
150
151/*
152 * This (illegal) pool name is used when temporarily importing a spa_t in order
153 * to get the vdev stats associated with the imported devices.
154 */
155#define	TRYIMPORT_NAME	"$import"
156
157/*
158 * ==========================================================================
159 * SPA properties routines
160 * ==========================================================================
161 */
162
163/*
164 * Add a (source=src, propname=propval) list to an nvlist.
165 */
166static void
167spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
168    uint64_t intval, zprop_source_t src)
169{
170	const char *propname = zpool_prop_to_name(prop);
171	nvlist_t *propval;
172
173	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
174	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
175
176	if (strval != NULL)
177		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
178	else
179		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
180
181	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
182	nvlist_free(propval);
183}
184
185/*
186 * Get property values from the spa configuration.
187 */
188static void
189spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
190{
191	vdev_t *rvd = spa->spa_root_vdev;
192	dsl_pool_t *pool = spa->spa_dsl_pool;
193	uint64_t size;
194	uint64_t alloc;
195	uint64_t space;
196	uint64_t cap, version;
197	zprop_source_t src = ZPROP_SRC_NONE;
198	spa_config_dirent_t *dp;
199
200	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
201
202	if (rvd != NULL) {
203		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
204		size = metaslab_class_get_space(spa_normal_class(spa));
205		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
206		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
207		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
208		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
209		    size - alloc, src);
210
211		space = 0;
212		for (int c = 0; c < rvd->vdev_children; c++) {
213			vdev_t *tvd = rvd->vdev_child[c];
214			space += tvd->vdev_max_asize - tvd->vdev_asize;
215		}
216		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
217		    src);
218
219		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
220		    (spa_mode(spa) == FREAD), src);
221
222		cap = (size == 0) ? 0 : (alloc * 100 / size);
223		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
224
225		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
226		    ddt_get_pool_dedup_ratio(spa), src);
227
228		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
229		    rvd->vdev_state, src);
230
231		version = spa_version(spa);
232		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
233			src = ZPROP_SRC_DEFAULT;
234		else
235			src = ZPROP_SRC_LOCAL;
236		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
237	}
238
239	if (pool != NULL) {
240		dsl_dir_t *freedir = pool->dp_free_dir;
241
242		/*
243		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
244		 * when opening pools before this version freedir will be NULL.
245		 */
246		if (freedir != NULL) {
247			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
248			    freedir->dd_phys->dd_used_bytes, src);
249		} else {
250			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
251			    NULL, 0, src);
252		}
253	}
254
255	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
256
257	if (spa->spa_comment != NULL) {
258		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
259		    0, ZPROP_SRC_LOCAL);
260	}
261
262	if (spa->spa_root != NULL)
263		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
264		    0, ZPROP_SRC_LOCAL);
265
266	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
267		if (dp->scd_path == NULL) {
268			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
269			    "none", 0, ZPROP_SRC_LOCAL);
270		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
271			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
272			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
273		}
274	}
275}
276
277/*
278 * Get zpool property values.
279 */
280int
281spa_prop_get(spa_t *spa, nvlist_t **nvp)
282{
283	objset_t *mos = spa->spa_meta_objset;
284	zap_cursor_t zc;
285	zap_attribute_t za;
286	int err;
287
288	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
289
290	mutex_enter(&spa->spa_props_lock);
291
292	/*
293	 * Get properties from the spa config.
294	 */
295	spa_prop_get_config(spa, nvp);
296
297	/* If no pool property object, no more prop to get. */
298	if (mos == NULL || spa->spa_pool_props_object == 0) {
299		mutex_exit(&spa->spa_props_lock);
300		return (0);
301	}
302
303	/*
304	 * Get properties from the MOS pool property object.
305	 */
306	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
307	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
308	    zap_cursor_advance(&zc)) {
309		uint64_t intval = 0;
310		char *strval = NULL;
311		zprop_source_t src = ZPROP_SRC_DEFAULT;
312		zpool_prop_t prop;
313
314		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
315			continue;
316
317		switch (za.za_integer_length) {
318		case 8:
319			/* integer property */
320			if (za.za_first_integer !=
321			    zpool_prop_default_numeric(prop))
322				src = ZPROP_SRC_LOCAL;
323
324			if (prop == ZPOOL_PROP_BOOTFS) {
325				dsl_pool_t *dp;
326				dsl_dataset_t *ds = NULL;
327
328				dp = spa_get_dsl(spa);
329				dsl_pool_config_enter(dp, FTAG);
330				if (err = dsl_dataset_hold_obj(dp,
331				    za.za_first_integer, FTAG, &ds)) {
332					dsl_pool_config_exit(dp, FTAG);
333					break;
334				}
335
336				strval = kmem_alloc(
337				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
338				    KM_SLEEP);
339				dsl_dataset_name(ds, strval);
340				dsl_dataset_rele(ds, FTAG);
341				dsl_pool_config_exit(dp, FTAG);
342			} else {
343				strval = NULL;
344				intval = za.za_first_integer;
345			}
346
347			spa_prop_add_list(*nvp, prop, strval, intval, src);
348
349			if (strval != NULL)
350				kmem_free(strval,
351				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
352
353			break;
354
355		case 1:
356			/* string property */
357			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
358			err = zap_lookup(mos, spa->spa_pool_props_object,
359			    za.za_name, 1, za.za_num_integers, strval);
360			if (err) {
361				kmem_free(strval, za.za_num_integers);
362				break;
363			}
364			spa_prop_add_list(*nvp, prop, strval, 0, src);
365			kmem_free(strval, za.za_num_integers);
366			break;
367
368		default:
369			break;
370		}
371	}
372	zap_cursor_fini(&zc);
373	mutex_exit(&spa->spa_props_lock);
374out:
375	if (err && err != ENOENT) {
376		nvlist_free(*nvp);
377		*nvp = NULL;
378		return (err);
379	}
380
381	return (0);
382}
383
384/*
385 * Validate the given pool properties nvlist and modify the list
386 * for the property values to be set.
387 */
388static int
389spa_prop_validate(spa_t *spa, nvlist_t *props)
390{
391	nvpair_t *elem;
392	int error = 0, reset_bootfs = 0;
393	uint64_t objnum = 0;
394	boolean_t has_feature = B_FALSE;
395
396	elem = NULL;
397	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
398		uint64_t intval;
399		char *strval, *slash, *check, *fname;
400		const char *propname = nvpair_name(elem);
401		zpool_prop_t prop = zpool_name_to_prop(propname);
402
403		switch (prop) {
404		case ZPROP_INVAL:
405			if (!zpool_prop_feature(propname)) {
406				error = SET_ERROR(EINVAL);
407				break;
408			}
409
410			/*
411			 * Sanitize the input.
412			 */
413			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
414				error = SET_ERROR(EINVAL);
415				break;
416			}
417
418			if (nvpair_value_uint64(elem, &intval) != 0) {
419				error = SET_ERROR(EINVAL);
420				break;
421			}
422
423			if (intval != 0) {
424				error = SET_ERROR(EINVAL);
425				break;
426			}
427
428			fname = strchr(propname, '@') + 1;
429			if (zfeature_lookup_name(fname, NULL) != 0) {
430				error = SET_ERROR(EINVAL);
431				break;
432			}
433
434			has_feature = B_TRUE;
435			break;
436
437		case ZPOOL_PROP_VERSION:
438			error = nvpair_value_uint64(elem, &intval);
439			if (!error &&
440			    (intval < spa_version(spa) ||
441			    intval > SPA_VERSION_BEFORE_FEATURES ||
442			    has_feature))
443				error = SET_ERROR(EINVAL);
444			break;
445
446		case ZPOOL_PROP_DELEGATION:
447		case ZPOOL_PROP_AUTOREPLACE:
448		case ZPOOL_PROP_LISTSNAPS:
449		case ZPOOL_PROP_AUTOEXPAND:
450			error = nvpair_value_uint64(elem, &intval);
451			if (!error && intval > 1)
452				error = SET_ERROR(EINVAL);
453			break;
454
455		case ZPOOL_PROP_BOOTFS:
456			/*
457			 * If the pool version is less than SPA_VERSION_BOOTFS,
458			 * or the pool is still being created (version == 0),
459			 * the bootfs property cannot be set.
460			 */
461			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
462				error = SET_ERROR(ENOTSUP);
463				break;
464			}
465
466			/*
467			 * Make sure the vdev config is bootable
468			 */
469			if (!vdev_is_bootable(spa->spa_root_vdev)) {
470				error = SET_ERROR(ENOTSUP);
471				break;
472			}
473
474			reset_bootfs = 1;
475
476			error = nvpair_value_string(elem, &strval);
477
478			if (!error) {
479				objset_t *os;
480				uint64_t compress;
481
482				if (strval == NULL || strval[0] == '\0') {
483					objnum = zpool_prop_default_numeric(
484					    ZPOOL_PROP_BOOTFS);
485					break;
486				}
487
488				if (error = dmu_objset_hold(strval, FTAG, &os))
489					break;
490
491				/* Must be ZPL and not gzip compressed. */
492
493				if (dmu_objset_type(os) != DMU_OST_ZFS) {
494					error = SET_ERROR(ENOTSUP);
495				} else if ((error =
496				    dsl_prop_get_int_ds(dmu_objset_ds(os),
497				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
498				    &compress)) == 0 &&
499				    !BOOTFS_COMPRESS_VALID(compress)) {
500					error = SET_ERROR(ENOTSUP);
501				} else {
502					objnum = dmu_objset_id(os);
503				}
504				dmu_objset_rele(os, FTAG);
505			}
506			break;
507
508		case ZPOOL_PROP_FAILUREMODE:
509			error = nvpair_value_uint64(elem, &intval);
510			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
511			    intval > ZIO_FAILURE_MODE_PANIC))
512				error = SET_ERROR(EINVAL);
513
514			/*
515			 * This is a special case which only occurs when
516			 * the pool has completely failed. This allows
517			 * the user to change the in-core failmode property
518			 * without syncing it out to disk (I/Os might
519			 * currently be blocked). We do this by returning
520			 * EIO to the caller (spa_prop_set) to trick it
521			 * into thinking we encountered a property validation
522			 * error.
523			 */
524			if (!error && spa_suspended(spa)) {
525				spa->spa_failmode = intval;
526				error = SET_ERROR(EIO);
527			}
528			break;
529
530		case ZPOOL_PROP_CACHEFILE:
531			if ((error = nvpair_value_string(elem, &strval)) != 0)
532				break;
533
534			if (strval[0] == '\0')
535				break;
536
537			if (strcmp(strval, "none") == 0)
538				break;
539
540			if (strval[0] != '/') {
541				error = SET_ERROR(EINVAL);
542				break;
543			}
544
545			slash = strrchr(strval, '/');
546			ASSERT(slash != NULL);
547
548			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
549			    strcmp(slash, "/..") == 0)
550				error = SET_ERROR(EINVAL);
551			break;
552
553		case ZPOOL_PROP_COMMENT:
554			if ((error = nvpair_value_string(elem, &strval)) != 0)
555				break;
556			for (check = strval; *check != '\0'; check++) {
557				/*
558				 * The kernel doesn't have an easy isprint()
559				 * check.  For this kernel check, we merely
560				 * check ASCII apart from DEL.  Fix this if
561				 * there is an easy-to-use kernel isprint().
562				 */
563				if (*check >= 0x7f) {
564					error = SET_ERROR(EINVAL);
565					break;
566				}
567				check++;
568			}
569			if (strlen(strval) > ZPROP_MAX_COMMENT)
570				error = E2BIG;
571			break;
572
573		case ZPOOL_PROP_DEDUPDITTO:
574			if (spa_version(spa) < SPA_VERSION_DEDUP)
575				error = SET_ERROR(ENOTSUP);
576			else
577				error = nvpair_value_uint64(elem, &intval);
578			if (error == 0 &&
579			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
580				error = SET_ERROR(EINVAL);
581			break;
582		}
583
584		if (error)
585			break;
586	}
587
588	if (!error && reset_bootfs) {
589		error = nvlist_remove(props,
590		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
591
592		if (!error) {
593			error = nvlist_add_uint64(props,
594			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
595		}
596	}
597
598	return (error);
599}
600
601void
602spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
603{
604	char *cachefile;
605	spa_config_dirent_t *dp;
606
607	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
608	    &cachefile) != 0)
609		return;
610
611	dp = kmem_alloc(sizeof (spa_config_dirent_t),
612	    KM_SLEEP);
613
614	if (cachefile[0] == '\0')
615		dp->scd_path = spa_strdup(spa_config_path);
616	else if (strcmp(cachefile, "none") == 0)
617		dp->scd_path = NULL;
618	else
619		dp->scd_path = spa_strdup(cachefile);
620
621	list_insert_head(&spa->spa_config_list, dp);
622	if (need_sync)
623		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
624}
625
626int
627spa_prop_set(spa_t *spa, nvlist_t *nvp)
628{
629	int error;
630	nvpair_t *elem = NULL;
631	boolean_t need_sync = B_FALSE;
632
633	if ((error = spa_prop_validate(spa, nvp)) != 0)
634		return (error);
635
636	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
637		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
638
639		if (prop == ZPOOL_PROP_CACHEFILE ||
640		    prop == ZPOOL_PROP_ALTROOT ||
641		    prop == ZPOOL_PROP_READONLY)
642			continue;
643
644		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
645			uint64_t ver;
646
647			if (prop == ZPOOL_PROP_VERSION) {
648				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
649			} else {
650				ASSERT(zpool_prop_feature(nvpair_name(elem)));
651				ver = SPA_VERSION_FEATURES;
652				need_sync = B_TRUE;
653			}
654
655			/* Save time if the version is already set. */
656			if (ver == spa_version(spa))
657				continue;
658
659			/*
660			 * In addition to the pool directory object, we might
661			 * create the pool properties object, the features for
662			 * read object, the features for write object, or the
663			 * feature descriptions object.
664			 */
665			error = dsl_sync_task(spa->spa_name, NULL,
666			    spa_sync_version, &ver, 6);
667			if (error)
668				return (error);
669			continue;
670		}
671
672		need_sync = B_TRUE;
673		break;
674	}
675
676	if (need_sync) {
677		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
678		    nvp, 6));
679	}
680
681	return (0);
682}
683
684/*
685 * If the bootfs property value is dsobj, clear it.
686 */
687void
688spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
689{
690	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
691		VERIFY(zap_remove(spa->spa_meta_objset,
692		    spa->spa_pool_props_object,
693		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
694		spa->spa_bootfs = 0;
695	}
696}
697
698/*ARGSUSED*/
699static int
700spa_change_guid_check(void *arg, dmu_tx_t *tx)
701{
702	uint64_t *newguid = arg;
703	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
704	vdev_t *rvd = spa->spa_root_vdev;
705	uint64_t vdev_state;
706
707	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
708	vdev_state = rvd->vdev_state;
709	spa_config_exit(spa, SCL_STATE, FTAG);
710
711	if (vdev_state != VDEV_STATE_HEALTHY)
712		return (SET_ERROR(ENXIO));
713
714	ASSERT3U(spa_guid(spa), !=, *newguid);
715
716	return (0);
717}
718
719static void
720spa_change_guid_sync(void *arg, dmu_tx_t *tx)
721{
722	uint64_t *newguid = arg;
723	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
724	uint64_t oldguid;
725	vdev_t *rvd = spa->spa_root_vdev;
726
727	oldguid = spa_guid(spa);
728
729	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
730	rvd->vdev_guid = *newguid;
731	rvd->vdev_guid_sum += (*newguid - oldguid);
732	vdev_config_dirty(rvd);
733	spa_config_exit(spa, SCL_STATE, FTAG);
734
735	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
736	    oldguid, *newguid);
737}
738
739/*
740 * Change the GUID for the pool.  This is done so that we can later
741 * re-import a pool built from a clone of our own vdevs.  We will modify
742 * the root vdev's guid, our own pool guid, and then mark all of our
743 * vdevs dirty.  Note that we must make sure that all our vdevs are
744 * online when we do this, or else any vdevs that weren't present
745 * would be orphaned from our pool.  We are also going to issue a
746 * sysevent to update any watchers.
747 */
748int
749spa_change_guid(spa_t *spa)
750{
751	int error;
752	uint64_t guid;
753
754	mutex_enter(&spa_namespace_lock);
755	guid = spa_generate_guid(NULL);
756
757	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
758	    spa_change_guid_sync, &guid, 5);
759
760	if (error == 0) {
761		spa_config_sync(spa, B_FALSE, B_TRUE);
762		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
763	}
764
765	mutex_exit(&spa_namespace_lock);
766
767	return (error);
768}
769
770/*
771 * ==========================================================================
772 * SPA state manipulation (open/create/destroy/import/export)
773 * ==========================================================================
774 */
775
776static int
777spa_error_entry_compare(const void *a, const void *b)
778{
779	spa_error_entry_t *sa = (spa_error_entry_t *)a;
780	spa_error_entry_t *sb = (spa_error_entry_t *)b;
781	int ret;
782
783	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
784	    sizeof (zbookmark_t));
785
786	if (ret < 0)
787		return (-1);
788	else if (ret > 0)
789		return (1);
790	else
791		return (0);
792}
793
794/*
795 * Utility function which retrieves copies of the current logs and
796 * re-initializes them in the process.
797 */
798void
799spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
800{
801	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
802
803	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
804	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
805
806	avl_create(&spa->spa_errlist_scrub,
807	    spa_error_entry_compare, sizeof (spa_error_entry_t),
808	    offsetof(spa_error_entry_t, se_avl));
809	avl_create(&spa->spa_errlist_last,
810	    spa_error_entry_compare, sizeof (spa_error_entry_t),
811	    offsetof(spa_error_entry_t, se_avl));
812}
813
814static void
815spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
816{
817	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
818	enum zti_modes mode = ztip->zti_mode;
819	uint_t value = ztip->zti_value;
820	uint_t count = ztip->zti_count;
821	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
822	char name[32];
823	uint_t flags = 0;
824	boolean_t batch = B_FALSE;
825
826	if (mode == ZTI_MODE_NULL) {
827		tqs->stqs_count = 0;
828		tqs->stqs_taskq = NULL;
829		return;
830	}
831
832	ASSERT3U(count, >, 0);
833
834	tqs->stqs_count = count;
835	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
836
837	for (uint_t i = 0; i < count; i++) {
838		taskq_t *tq;
839
840		switch (mode) {
841		case ZTI_MODE_FIXED:
842			ASSERT3U(value, >=, 1);
843			value = MAX(value, 1);
844			break;
845
846		case ZTI_MODE_BATCH:
847			batch = B_TRUE;
848			flags |= TASKQ_THREADS_CPU_PCT;
849			value = zio_taskq_batch_pct;
850			break;
851
852		case ZTI_MODE_ONLINE_PERCENT:
853			flags |= TASKQ_THREADS_CPU_PCT;
854			break;
855
856		default:
857			panic("unrecognized mode for %s_%s taskq (%u:%u) in "
858			    "spa_activate()",
859			    zio_type_name[t], zio_taskq_types[q], mode, value);
860			break;
861		}
862
863		if (count > 1) {
864			(void) snprintf(name, sizeof (name), "%s_%s_%u",
865			    zio_type_name[t], zio_taskq_types[q], i);
866		} else {
867			(void) snprintf(name, sizeof (name), "%s_%s",
868			    zio_type_name[t], zio_taskq_types[q]);
869		}
870
871		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
872			if (batch)
873				flags |= TASKQ_DC_BATCH;
874
875			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
876			    spa->spa_proc, zio_taskq_basedc, flags);
877		} else {
878			tq = taskq_create_proc(name, value, maxclsyspri, 50,
879			    INT_MAX, spa->spa_proc, flags);
880		}
881
882		tqs->stqs_taskq[i] = tq;
883	}
884}
885
886static void
887spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
888{
889	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
890
891	if (tqs->stqs_taskq == NULL) {
892		ASSERT0(tqs->stqs_count);
893		return;
894	}
895
896	for (uint_t i = 0; i < tqs->stqs_count; i++) {
897		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
898		taskq_destroy(tqs->stqs_taskq[i]);
899	}
900
901	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
902	tqs->stqs_taskq = NULL;
903}
904
905/*
906 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
907 * Note that a type may have multiple discrete taskqs to avoid lock contention
908 * on the taskq itself. In that case we choose which taskq at random by using
909 * the low bits of gethrtime().
910 */
911void
912spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
913    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
914{
915	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
916	taskq_t *tq;
917
918	ASSERT3P(tqs->stqs_taskq, !=, NULL);
919	ASSERT3U(tqs->stqs_count, !=, 0);
920
921	if (tqs->stqs_count == 1) {
922		tq = tqs->stqs_taskq[0];
923	} else {
924		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
925	}
926
927	taskq_dispatch_ent(tq, func, arg, flags, ent);
928}
929
930static void
931spa_create_zio_taskqs(spa_t *spa)
932{
933	for (int t = 0; t < ZIO_TYPES; t++) {
934		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
935			spa_taskqs_init(spa, t, q);
936		}
937	}
938}
939
940#ifdef _KERNEL
941static void
942spa_thread(void *arg)
943{
944	callb_cpr_t cprinfo;
945
946	spa_t *spa = arg;
947	user_t *pu = PTOU(curproc);
948
949	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
950	    spa->spa_name);
951
952	ASSERT(curproc != &p0);
953	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
954	    "zpool-%s", spa->spa_name);
955	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
956
957	/* bind this thread to the requested psrset */
958	if (zio_taskq_psrset_bind != PS_NONE) {
959		pool_lock();
960		mutex_enter(&cpu_lock);
961		mutex_enter(&pidlock);
962		mutex_enter(&curproc->p_lock);
963
964		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
965		    0, NULL, NULL) == 0)  {
966			curthread->t_bind_pset = zio_taskq_psrset_bind;
967		} else {
968			cmn_err(CE_WARN,
969			    "Couldn't bind process for zfs pool \"%s\" to "
970			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
971		}
972
973		mutex_exit(&curproc->p_lock);
974		mutex_exit(&pidlock);
975		mutex_exit(&cpu_lock);
976		pool_unlock();
977	}
978
979	if (zio_taskq_sysdc) {
980		sysdc_thread_enter(curthread, 100, 0);
981	}
982
983	spa->spa_proc = curproc;
984	spa->spa_did = curthread->t_did;
985
986	spa_create_zio_taskqs(spa);
987
988	mutex_enter(&spa->spa_proc_lock);
989	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
990
991	spa->spa_proc_state = SPA_PROC_ACTIVE;
992	cv_broadcast(&spa->spa_proc_cv);
993
994	CALLB_CPR_SAFE_BEGIN(&cprinfo);
995	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
996		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
997	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
998
999	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1000	spa->spa_proc_state = SPA_PROC_GONE;
1001	spa->spa_proc = &p0;
1002	cv_broadcast(&spa->spa_proc_cv);
1003	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1004
1005	mutex_enter(&curproc->p_lock);
1006	lwp_exit();
1007}
1008#endif
1009
1010/*
1011 * Activate an uninitialized pool.
1012 */
1013static void
1014spa_activate(spa_t *spa, int mode)
1015{
1016	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1017
1018	spa->spa_state = POOL_STATE_ACTIVE;
1019	spa->spa_mode = mode;
1020
1021	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1022	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1023
1024	/* Try to create a covering process */
1025	mutex_enter(&spa->spa_proc_lock);
1026	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1027	ASSERT(spa->spa_proc == &p0);
1028	spa->spa_did = 0;
1029
1030	/* Only create a process if we're going to be around a while. */
1031	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1032		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1033		    NULL, 0) == 0) {
1034			spa->spa_proc_state = SPA_PROC_CREATED;
1035			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1036				cv_wait(&spa->spa_proc_cv,
1037				    &spa->spa_proc_lock);
1038			}
1039			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1040			ASSERT(spa->spa_proc != &p0);
1041			ASSERT(spa->spa_did != 0);
1042		} else {
1043#ifdef _KERNEL
1044			cmn_err(CE_WARN,
1045			    "Couldn't create process for zfs pool \"%s\"\n",
1046			    spa->spa_name);
1047#endif
1048		}
1049	}
1050	mutex_exit(&spa->spa_proc_lock);
1051
1052	/* If we didn't create a process, we need to create our taskqs. */
1053	if (spa->spa_proc == &p0) {
1054		spa_create_zio_taskqs(spa);
1055	}
1056
1057	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1058	    offsetof(vdev_t, vdev_config_dirty_node));
1059	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1060	    offsetof(vdev_t, vdev_state_dirty_node));
1061
1062	txg_list_create(&spa->spa_vdev_txg_list,
1063	    offsetof(struct vdev, vdev_txg_node));
1064
1065	avl_create(&spa->spa_errlist_scrub,
1066	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1067	    offsetof(spa_error_entry_t, se_avl));
1068	avl_create(&spa->spa_errlist_last,
1069	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1070	    offsetof(spa_error_entry_t, se_avl));
1071}
1072
1073/*
1074 * Opposite of spa_activate().
1075 */
1076static void
1077spa_deactivate(spa_t *spa)
1078{
1079	ASSERT(spa->spa_sync_on == B_FALSE);
1080	ASSERT(spa->spa_dsl_pool == NULL);
1081	ASSERT(spa->spa_root_vdev == NULL);
1082	ASSERT(spa->spa_async_zio_root == NULL);
1083	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1084
1085	txg_list_destroy(&spa->spa_vdev_txg_list);
1086
1087	list_destroy(&spa->spa_config_dirty_list);
1088	list_destroy(&spa->spa_state_dirty_list);
1089
1090	for (int t = 0; t < ZIO_TYPES; t++) {
1091		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1092			spa_taskqs_fini(spa, t, q);
1093		}
1094	}
1095
1096	metaslab_class_destroy(spa->spa_normal_class);
1097	spa->spa_normal_class = NULL;
1098
1099	metaslab_class_destroy(spa->spa_log_class);
1100	spa->spa_log_class = NULL;
1101
1102	/*
1103	 * If this was part of an import or the open otherwise failed, we may
1104	 * still have errors left in the queues.  Empty them just in case.
1105	 */
1106	spa_errlog_drain(spa);
1107
1108	avl_destroy(&spa->spa_errlist_scrub);
1109	avl_destroy(&spa->spa_errlist_last);
1110
1111	spa->spa_state = POOL_STATE_UNINITIALIZED;
1112
1113	mutex_enter(&spa->spa_proc_lock);
1114	if (spa->spa_proc_state != SPA_PROC_NONE) {
1115		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1116		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1117		cv_broadcast(&spa->spa_proc_cv);
1118		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1119			ASSERT(spa->spa_proc != &p0);
1120			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1121		}
1122		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1123		spa->spa_proc_state = SPA_PROC_NONE;
1124	}
1125	ASSERT(spa->spa_proc == &p0);
1126	mutex_exit(&spa->spa_proc_lock);
1127
1128	/*
1129	 * We want to make sure spa_thread() has actually exited the ZFS
1130	 * module, so that the module can't be unloaded out from underneath
1131	 * it.
1132	 */
1133	if (spa->spa_did != 0) {
1134		thread_join(spa->spa_did);
1135		spa->spa_did = 0;
1136	}
1137}
1138
1139/*
1140 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1141 * will create all the necessary vdevs in the appropriate layout, with each vdev
1142 * in the CLOSED state.  This will prep the pool before open/creation/import.
1143 * All vdev validation is done by the vdev_alloc() routine.
1144 */
1145static int
1146spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1147    uint_t id, int atype)
1148{
1149	nvlist_t **child;
1150	uint_t children;
1151	int error;
1152
1153	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1154		return (error);
1155
1156	if ((*vdp)->vdev_ops->vdev_op_leaf)
1157		return (0);
1158
1159	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1160	    &child, &children);
1161
1162	if (error == ENOENT)
1163		return (0);
1164
1165	if (error) {
1166		vdev_free(*vdp);
1167		*vdp = NULL;
1168		return (SET_ERROR(EINVAL));
1169	}
1170
1171	for (int c = 0; c < children; c++) {
1172		vdev_t *vd;
1173		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1174		    atype)) != 0) {
1175			vdev_free(*vdp);
1176			*vdp = NULL;
1177			return (error);
1178		}
1179	}
1180
1181	ASSERT(*vdp != NULL);
1182
1183	return (0);
1184}
1185
1186/*
1187 * Opposite of spa_load().
1188 */
1189static void
1190spa_unload(spa_t *spa)
1191{
1192	int i;
1193
1194	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1195
1196	/*
1197	 * Stop async tasks.
1198	 */
1199	spa_async_suspend(spa);
1200
1201	/*
1202	 * Stop syncing.
1203	 */
1204	if (spa->spa_sync_on) {
1205		txg_sync_stop(spa->spa_dsl_pool);
1206		spa->spa_sync_on = B_FALSE;
1207	}
1208
1209	/*
1210	 * Wait for any outstanding async I/O to complete.
1211	 */
1212	if (spa->spa_async_zio_root != NULL) {
1213		(void) zio_wait(spa->spa_async_zio_root);
1214		spa->spa_async_zio_root = NULL;
1215	}
1216
1217	bpobj_close(&spa->spa_deferred_bpobj);
1218
1219	/*
1220	 * Close the dsl pool.
1221	 */
1222	if (spa->spa_dsl_pool) {
1223		dsl_pool_close(spa->spa_dsl_pool);
1224		spa->spa_dsl_pool = NULL;
1225		spa->spa_meta_objset = NULL;
1226	}
1227
1228	ddt_unload(spa);
1229
1230	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1231
1232	/*
1233	 * Drop and purge level 2 cache
1234	 */
1235	spa_l2cache_drop(spa);
1236
1237	/*
1238	 * Close all vdevs.
1239	 */
1240	if (spa->spa_root_vdev)
1241		vdev_free(spa->spa_root_vdev);
1242	ASSERT(spa->spa_root_vdev == NULL);
1243
1244	for (i = 0; i < spa->spa_spares.sav_count; i++)
1245		vdev_free(spa->spa_spares.sav_vdevs[i]);
1246	if (spa->spa_spares.sav_vdevs) {
1247		kmem_free(spa->spa_spares.sav_vdevs,
1248		    spa->spa_spares.sav_count * sizeof (void *));
1249		spa->spa_spares.sav_vdevs = NULL;
1250	}
1251	if (spa->spa_spares.sav_config) {
1252		nvlist_free(spa->spa_spares.sav_config);
1253		spa->spa_spares.sav_config = NULL;
1254	}
1255	spa->spa_spares.sav_count = 0;
1256
1257	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1258		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1259		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1260	}
1261	if (spa->spa_l2cache.sav_vdevs) {
1262		kmem_free(spa->spa_l2cache.sav_vdevs,
1263		    spa->spa_l2cache.sav_count * sizeof (void *));
1264		spa->spa_l2cache.sav_vdevs = NULL;
1265	}
1266	if (spa->spa_l2cache.sav_config) {
1267		nvlist_free(spa->spa_l2cache.sav_config);
1268		spa->spa_l2cache.sav_config = NULL;
1269	}
1270	spa->spa_l2cache.sav_count = 0;
1271
1272	spa->spa_async_suspended = 0;
1273
1274	if (spa->spa_comment != NULL) {
1275		spa_strfree(spa->spa_comment);
1276		spa->spa_comment = NULL;
1277	}
1278
1279	spa_config_exit(spa, SCL_ALL, FTAG);
1280}
1281
1282/*
1283 * Load (or re-load) the current list of vdevs describing the active spares for
1284 * this pool.  When this is called, we have some form of basic information in
1285 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1286 * then re-generate a more complete list including status information.
1287 */
1288static void
1289spa_load_spares(spa_t *spa)
1290{
1291	nvlist_t **spares;
1292	uint_t nspares;
1293	int i;
1294	vdev_t *vd, *tvd;
1295
1296	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1297
1298	/*
1299	 * First, close and free any existing spare vdevs.
1300	 */
1301	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1302		vd = spa->spa_spares.sav_vdevs[i];
1303
1304		/* Undo the call to spa_activate() below */
1305		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1306		    B_FALSE)) != NULL && tvd->vdev_isspare)
1307			spa_spare_remove(tvd);
1308		vdev_close(vd);
1309		vdev_free(vd);
1310	}
1311
1312	if (spa->spa_spares.sav_vdevs)
1313		kmem_free(spa->spa_spares.sav_vdevs,
1314		    spa->spa_spares.sav_count * sizeof (void *));
1315
1316	if (spa->spa_spares.sav_config == NULL)
1317		nspares = 0;
1318	else
1319		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1320		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1321
1322	spa->spa_spares.sav_count = (int)nspares;
1323	spa->spa_spares.sav_vdevs = NULL;
1324
1325	if (nspares == 0)
1326		return;
1327
1328	/*
1329	 * Construct the array of vdevs, opening them to get status in the
1330	 * process.   For each spare, there is potentially two different vdev_t
1331	 * structures associated with it: one in the list of spares (used only
1332	 * for basic validation purposes) and one in the active vdev
1333	 * configuration (if it's spared in).  During this phase we open and
1334	 * validate each vdev on the spare list.  If the vdev also exists in the
1335	 * active configuration, then we also mark this vdev as an active spare.
1336	 */
1337	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1338	    KM_SLEEP);
1339	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1340		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1341		    VDEV_ALLOC_SPARE) == 0);
1342		ASSERT(vd != NULL);
1343
1344		spa->spa_spares.sav_vdevs[i] = vd;
1345
1346		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1347		    B_FALSE)) != NULL) {
1348			if (!tvd->vdev_isspare)
1349				spa_spare_add(tvd);
1350
1351			/*
1352			 * We only mark the spare active if we were successfully
1353			 * able to load the vdev.  Otherwise, importing a pool
1354			 * with a bad active spare would result in strange
1355			 * behavior, because multiple pool would think the spare
1356			 * is actively in use.
1357			 *
1358			 * There is a vulnerability here to an equally bizarre
1359			 * circumstance, where a dead active spare is later
1360			 * brought back to life (onlined or otherwise).  Given
1361			 * the rarity of this scenario, and the extra complexity
1362			 * it adds, we ignore the possibility.
1363			 */
1364			if (!vdev_is_dead(tvd))
1365				spa_spare_activate(tvd);
1366		}
1367
1368		vd->vdev_top = vd;
1369		vd->vdev_aux = &spa->spa_spares;
1370
1371		if (vdev_open(vd) != 0)
1372			continue;
1373
1374		if (vdev_validate_aux(vd) == 0)
1375			spa_spare_add(vd);
1376	}
1377
1378	/*
1379	 * Recompute the stashed list of spares, with status information
1380	 * this time.
1381	 */
1382	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1383	    DATA_TYPE_NVLIST_ARRAY) == 0);
1384
1385	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1386	    KM_SLEEP);
1387	for (i = 0; i < spa->spa_spares.sav_count; i++)
1388		spares[i] = vdev_config_generate(spa,
1389		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1390	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1391	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1392	for (i = 0; i < spa->spa_spares.sav_count; i++)
1393		nvlist_free(spares[i]);
1394	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1395}
1396
1397/*
1398 * Load (or re-load) the current list of vdevs describing the active l2cache for
1399 * this pool.  When this is called, we have some form of basic information in
1400 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1401 * then re-generate a more complete list including status information.
1402 * Devices which are already active have their details maintained, and are
1403 * not re-opened.
1404 */
1405static void
1406spa_load_l2cache(spa_t *spa)
1407{
1408	nvlist_t **l2cache;
1409	uint_t nl2cache;
1410	int i, j, oldnvdevs;
1411	uint64_t guid;
1412	vdev_t *vd, **oldvdevs, **newvdevs;
1413	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1414
1415	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1416
1417	if (sav->sav_config != NULL) {
1418		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1419		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1420		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1421	} else {
1422		nl2cache = 0;
1423		newvdevs = NULL;
1424	}
1425
1426	oldvdevs = sav->sav_vdevs;
1427	oldnvdevs = sav->sav_count;
1428	sav->sav_vdevs = NULL;
1429	sav->sav_count = 0;
1430
1431	/*
1432	 * Process new nvlist of vdevs.
1433	 */
1434	for (i = 0; i < nl2cache; i++) {
1435		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1436		    &guid) == 0);
1437
1438		newvdevs[i] = NULL;
1439		for (j = 0; j < oldnvdevs; j++) {
1440			vd = oldvdevs[j];
1441			if (vd != NULL && guid == vd->vdev_guid) {
1442				/*
1443				 * Retain previous vdev for add/remove ops.
1444				 */
1445				newvdevs[i] = vd;
1446				oldvdevs[j] = NULL;
1447				break;
1448			}
1449		}
1450
1451		if (newvdevs[i] == NULL) {
1452			/*
1453			 * Create new vdev
1454			 */
1455			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1456			    VDEV_ALLOC_L2CACHE) == 0);
1457			ASSERT(vd != NULL);
1458			newvdevs[i] = vd;
1459
1460			/*
1461			 * Commit this vdev as an l2cache device,
1462			 * even if it fails to open.
1463			 */
1464			spa_l2cache_add(vd);
1465
1466			vd->vdev_top = vd;
1467			vd->vdev_aux = sav;
1468
1469			spa_l2cache_activate(vd);
1470
1471			if (vdev_open(vd) != 0)
1472				continue;
1473
1474			(void) vdev_validate_aux(vd);
1475
1476			if (!vdev_is_dead(vd))
1477				l2arc_add_vdev(spa, vd);
1478		}
1479	}
1480
1481	/*
1482	 * Purge vdevs that were dropped
1483	 */
1484	for (i = 0; i < oldnvdevs; i++) {
1485		uint64_t pool;
1486
1487		vd = oldvdevs[i];
1488		if (vd != NULL) {
1489			ASSERT(vd->vdev_isl2cache);
1490
1491			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1492			    pool != 0ULL && l2arc_vdev_present(vd))
1493				l2arc_remove_vdev(vd);
1494			vdev_clear_stats(vd);
1495			vdev_free(vd);
1496		}
1497	}
1498
1499	if (oldvdevs)
1500		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1501
1502	if (sav->sav_config == NULL)
1503		goto out;
1504
1505	sav->sav_vdevs = newvdevs;
1506	sav->sav_count = (int)nl2cache;
1507
1508	/*
1509	 * Recompute the stashed list of l2cache devices, with status
1510	 * information this time.
1511	 */
1512	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1513	    DATA_TYPE_NVLIST_ARRAY) == 0);
1514
1515	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1516	for (i = 0; i < sav->sav_count; i++)
1517		l2cache[i] = vdev_config_generate(spa,
1518		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1519	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1520	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1521out:
1522	for (i = 0; i < sav->sav_count; i++)
1523		nvlist_free(l2cache[i]);
1524	if (sav->sav_count)
1525		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1526}
1527
1528static int
1529load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1530{
1531	dmu_buf_t *db;
1532	char *packed = NULL;
1533	size_t nvsize = 0;
1534	int error;
1535	*value = NULL;
1536
1537	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1538	nvsize = *(uint64_t *)db->db_data;
1539	dmu_buf_rele(db, FTAG);
1540
1541	packed = kmem_alloc(nvsize, KM_SLEEP);
1542	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1543	    DMU_READ_PREFETCH);
1544	if (error == 0)
1545		error = nvlist_unpack(packed, nvsize, value, 0);
1546	kmem_free(packed, nvsize);
1547
1548	return (error);
1549}
1550
1551/*
1552 * Checks to see if the given vdev could not be opened, in which case we post a
1553 * sysevent to notify the autoreplace code that the device has been removed.
1554 */
1555static void
1556spa_check_removed(vdev_t *vd)
1557{
1558	for (int c = 0; c < vd->vdev_children; c++)
1559		spa_check_removed(vd->vdev_child[c]);
1560
1561	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1562	    !vd->vdev_ishole) {
1563		zfs_post_autoreplace(vd->vdev_spa, vd);
1564		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1565	}
1566}
1567
1568/*
1569 * Validate the current config against the MOS config
1570 */
1571static boolean_t
1572spa_config_valid(spa_t *spa, nvlist_t *config)
1573{
1574	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1575	nvlist_t *nv;
1576
1577	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1578
1579	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1580	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1581
1582	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1583
1584	/*
1585	 * If we're doing a normal import, then build up any additional
1586	 * diagnostic information about missing devices in this config.
1587	 * We'll pass this up to the user for further processing.
1588	 */
1589	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1590		nvlist_t **child, *nv;
1591		uint64_t idx = 0;
1592
1593		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1594		    KM_SLEEP);
1595		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1596
1597		for (int c = 0; c < rvd->vdev_children; c++) {
1598			vdev_t *tvd = rvd->vdev_child[c];
1599			vdev_t *mtvd  = mrvd->vdev_child[c];
1600
1601			if (tvd->vdev_ops == &vdev_missing_ops &&
1602			    mtvd->vdev_ops != &vdev_missing_ops &&
1603			    mtvd->vdev_islog)
1604				child[idx++] = vdev_config_generate(spa, mtvd,
1605				    B_FALSE, 0);
1606		}
1607
1608		if (idx) {
1609			VERIFY(nvlist_add_nvlist_array(nv,
1610			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1611			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1612			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1613
1614			for (int i = 0; i < idx; i++)
1615				nvlist_free(child[i]);
1616		}
1617		nvlist_free(nv);
1618		kmem_free(child, rvd->vdev_children * sizeof (char **));
1619	}
1620
1621	/*
1622	 * Compare the root vdev tree with the information we have
1623	 * from the MOS config (mrvd). Check each top-level vdev
1624	 * with the corresponding MOS config top-level (mtvd).
1625	 */
1626	for (int c = 0; c < rvd->vdev_children; c++) {
1627		vdev_t *tvd = rvd->vdev_child[c];
1628		vdev_t *mtvd  = mrvd->vdev_child[c];
1629
1630		/*
1631		 * Resolve any "missing" vdevs in the current configuration.
1632		 * If we find that the MOS config has more accurate information
1633		 * about the top-level vdev then use that vdev instead.
1634		 */
1635		if (tvd->vdev_ops == &vdev_missing_ops &&
1636		    mtvd->vdev_ops != &vdev_missing_ops) {
1637
1638			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1639				continue;
1640
1641			/*
1642			 * Device specific actions.
1643			 */
1644			if (mtvd->vdev_islog) {
1645				spa_set_log_state(spa, SPA_LOG_CLEAR);
1646			} else {
1647				/*
1648				 * XXX - once we have 'readonly' pool
1649				 * support we should be able to handle
1650				 * missing data devices by transitioning
1651				 * the pool to readonly.
1652				 */
1653				continue;
1654			}
1655
1656			/*
1657			 * Swap the missing vdev with the data we were
1658			 * able to obtain from the MOS config.
1659			 */
1660			vdev_remove_child(rvd, tvd);
1661			vdev_remove_child(mrvd, mtvd);
1662
1663			vdev_add_child(rvd, mtvd);
1664			vdev_add_child(mrvd, tvd);
1665
1666			spa_config_exit(spa, SCL_ALL, FTAG);
1667			vdev_load(mtvd);
1668			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1669
1670			vdev_reopen(rvd);
1671		} else if (mtvd->vdev_islog) {
1672			/*
1673			 * Load the slog device's state from the MOS config
1674			 * since it's possible that the label does not
1675			 * contain the most up-to-date information.
1676			 */
1677			vdev_load_log_state(tvd, mtvd);
1678			vdev_reopen(tvd);
1679		}
1680	}
1681	vdev_free(mrvd);
1682	spa_config_exit(spa, SCL_ALL, FTAG);
1683
1684	/*
1685	 * Ensure we were able to validate the config.
1686	 */
1687	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1688}
1689
1690/*
1691 * Check for missing log devices
1692 */
1693static boolean_t
1694spa_check_logs(spa_t *spa)
1695{
1696	boolean_t rv = B_FALSE;
1697
1698	switch (spa->spa_log_state) {
1699	case SPA_LOG_MISSING:
1700		/* need to recheck in case slog has been restored */
1701	case SPA_LOG_UNKNOWN:
1702		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1703		    NULL, DS_FIND_CHILDREN) != 0);
1704		if (rv)
1705			spa_set_log_state(spa, SPA_LOG_MISSING);
1706		break;
1707	}
1708	return (rv);
1709}
1710
1711static boolean_t
1712spa_passivate_log(spa_t *spa)
1713{
1714	vdev_t *rvd = spa->spa_root_vdev;
1715	boolean_t slog_found = B_FALSE;
1716
1717	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1718
1719	if (!spa_has_slogs(spa))
1720		return (B_FALSE);
1721
1722	for (int c = 0; c < rvd->vdev_children; c++) {
1723		vdev_t *tvd = rvd->vdev_child[c];
1724		metaslab_group_t *mg = tvd->vdev_mg;
1725
1726		if (tvd->vdev_islog) {
1727			metaslab_group_passivate(mg);
1728			slog_found = B_TRUE;
1729		}
1730	}
1731
1732	return (slog_found);
1733}
1734
1735static void
1736spa_activate_log(spa_t *spa)
1737{
1738	vdev_t *rvd = spa->spa_root_vdev;
1739
1740	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1741
1742	for (int c = 0; c < rvd->vdev_children; c++) {
1743		vdev_t *tvd = rvd->vdev_child[c];
1744		metaslab_group_t *mg = tvd->vdev_mg;
1745
1746		if (tvd->vdev_islog)
1747			metaslab_group_activate(mg);
1748	}
1749}
1750
1751int
1752spa_offline_log(spa_t *spa)
1753{
1754	int error;
1755
1756	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1757	    NULL, DS_FIND_CHILDREN);
1758	if (error == 0) {
1759		/*
1760		 * We successfully offlined the log device, sync out the
1761		 * current txg so that the "stubby" block can be removed
1762		 * by zil_sync().
1763		 */
1764		txg_wait_synced(spa->spa_dsl_pool, 0);
1765	}
1766	return (error);
1767}
1768
1769static void
1770spa_aux_check_removed(spa_aux_vdev_t *sav)
1771{
1772	for (int i = 0; i < sav->sav_count; i++)
1773		spa_check_removed(sav->sav_vdevs[i]);
1774}
1775
1776void
1777spa_claim_notify(zio_t *zio)
1778{
1779	spa_t *spa = zio->io_spa;
1780
1781	if (zio->io_error)
1782		return;
1783
1784	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1785	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1786		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1787	mutex_exit(&spa->spa_props_lock);
1788}
1789
1790typedef struct spa_load_error {
1791	uint64_t	sle_meta_count;
1792	uint64_t	sle_data_count;
1793} spa_load_error_t;
1794
1795static void
1796spa_load_verify_done(zio_t *zio)
1797{
1798	blkptr_t *bp = zio->io_bp;
1799	spa_load_error_t *sle = zio->io_private;
1800	dmu_object_type_t type = BP_GET_TYPE(bp);
1801	int error = zio->io_error;
1802
1803	if (error) {
1804		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1805		    type != DMU_OT_INTENT_LOG)
1806			atomic_add_64(&sle->sle_meta_count, 1);
1807		else
1808			atomic_add_64(&sle->sle_data_count, 1);
1809	}
1810	zio_data_buf_free(zio->io_data, zio->io_size);
1811}
1812
1813/*ARGSUSED*/
1814static int
1815spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1816    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1817{
1818	if (bp != NULL) {
1819		zio_t *rio = arg;
1820		size_t size = BP_GET_PSIZE(bp);
1821		void *data = zio_data_buf_alloc(size);
1822
1823		zio_nowait(zio_read(rio, spa, bp, data, size,
1824		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1825		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1826		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1827	}
1828	return (0);
1829}
1830
1831static int
1832spa_load_verify(spa_t *spa)
1833{
1834	zio_t *rio;
1835	spa_load_error_t sle = { 0 };
1836	zpool_rewind_policy_t policy;
1837	boolean_t verify_ok = B_FALSE;
1838	int error;
1839
1840	zpool_get_rewind_policy(spa->spa_config, &policy);
1841
1842	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1843		return (0);
1844
1845	rio = zio_root(spa, NULL, &sle,
1846	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1847
1848	error = traverse_pool(spa, spa->spa_verify_min_txg,
1849	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1850
1851	(void) zio_wait(rio);
1852
1853	spa->spa_load_meta_errors = sle.sle_meta_count;
1854	spa->spa_load_data_errors = sle.sle_data_count;
1855
1856	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1857	    sle.sle_data_count <= policy.zrp_maxdata) {
1858		int64_t loss = 0;
1859
1860		verify_ok = B_TRUE;
1861		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1862		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1863
1864		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1865		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1866		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1867		VERIFY(nvlist_add_int64(spa->spa_load_info,
1868		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1869		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1870		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1871	} else {
1872		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1873	}
1874
1875	if (error) {
1876		if (error != ENXIO && error != EIO)
1877			error = SET_ERROR(EIO);
1878		return (error);
1879	}
1880
1881	return (verify_ok ? 0 : EIO);
1882}
1883
1884/*
1885 * Find a value in the pool props object.
1886 */
1887static void
1888spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1889{
1890	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1891	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1892}
1893
1894/*
1895 * Find a value in the pool directory object.
1896 */
1897static int
1898spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1899{
1900	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1901	    name, sizeof (uint64_t), 1, val));
1902}
1903
1904static int
1905spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1906{
1907	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1908	return (err);
1909}
1910
1911/*
1912 * Fix up config after a partly-completed split.  This is done with the
1913 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1914 * pool have that entry in their config, but only the splitting one contains
1915 * a list of all the guids of the vdevs that are being split off.
1916 *
1917 * This function determines what to do with that list: either rejoin
1918 * all the disks to the pool, or complete the splitting process.  To attempt
1919 * the rejoin, each disk that is offlined is marked online again, and
1920 * we do a reopen() call.  If the vdev label for every disk that was
1921 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1922 * then we call vdev_split() on each disk, and complete the split.
1923 *
1924 * Otherwise we leave the config alone, with all the vdevs in place in
1925 * the original pool.
1926 */
1927static void
1928spa_try_repair(spa_t *spa, nvlist_t *config)
1929{
1930	uint_t extracted;
1931	uint64_t *glist;
1932	uint_t i, gcount;
1933	nvlist_t *nvl;
1934	vdev_t **vd;
1935	boolean_t attempt_reopen;
1936
1937	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1938		return;
1939
1940	/* check that the config is complete */
1941	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1942	    &glist, &gcount) != 0)
1943		return;
1944
1945	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1946
1947	/* attempt to online all the vdevs & validate */
1948	attempt_reopen = B_TRUE;
1949	for (i = 0; i < gcount; i++) {
1950		if (glist[i] == 0)	/* vdev is hole */
1951			continue;
1952
1953		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1954		if (vd[i] == NULL) {
1955			/*
1956			 * Don't bother attempting to reopen the disks;
1957			 * just do the split.
1958			 */
1959			attempt_reopen = B_FALSE;
1960		} else {
1961			/* attempt to re-online it */
1962			vd[i]->vdev_offline = B_FALSE;
1963		}
1964	}
1965
1966	if (attempt_reopen) {
1967		vdev_reopen(spa->spa_root_vdev);
1968
1969		/* check each device to see what state it's in */
1970		for (extracted = 0, i = 0; i < gcount; i++) {
1971			if (vd[i] != NULL &&
1972			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1973				break;
1974			++extracted;
1975		}
1976	}
1977
1978	/*
1979	 * If every disk has been moved to the new pool, or if we never
1980	 * even attempted to look at them, then we split them off for
1981	 * good.
1982	 */
1983	if (!attempt_reopen || gcount == extracted) {
1984		for (i = 0; i < gcount; i++)
1985			if (vd[i] != NULL)
1986				vdev_split(vd[i]);
1987		vdev_reopen(spa->spa_root_vdev);
1988	}
1989
1990	kmem_free(vd, gcount * sizeof (vdev_t *));
1991}
1992
1993static int
1994spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1995    boolean_t mosconfig)
1996{
1997	nvlist_t *config = spa->spa_config;
1998	char *ereport = FM_EREPORT_ZFS_POOL;
1999	char *comment;
2000	int error;
2001	uint64_t pool_guid;
2002	nvlist_t *nvl;
2003
2004	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2005		return (SET_ERROR(EINVAL));
2006
2007	ASSERT(spa->spa_comment == NULL);
2008	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2009		spa->spa_comment = spa_strdup(comment);
2010
2011	/*
2012	 * Versioning wasn't explicitly added to the label until later, so if
2013	 * it's not present treat it as the initial version.
2014	 */
2015	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2016	    &spa->spa_ubsync.ub_version) != 0)
2017		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2018
2019	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2020	    &spa->spa_config_txg);
2021
2022	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2023	    spa_guid_exists(pool_guid, 0)) {
2024		error = SET_ERROR(EEXIST);
2025	} else {
2026		spa->spa_config_guid = pool_guid;
2027
2028		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2029		    &nvl) == 0) {
2030			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2031			    KM_SLEEP) == 0);
2032		}
2033
2034		nvlist_free(spa->spa_load_info);
2035		spa->spa_load_info = fnvlist_alloc();
2036
2037		gethrestime(&spa->spa_loaded_ts);
2038		error = spa_load_impl(spa, pool_guid, config, state, type,
2039		    mosconfig, &ereport);
2040	}
2041
2042	spa->spa_minref = refcount_count(&spa->spa_refcount);
2043	if (error) {
2044		if (error != EEXIST) {
2045			spa->spa_loaded_ts.tv_sec = 0;
2046			spa->spa_loaded_ts.tv_nsec = 0;
2047		}
2048		if (error != EBADF) {
2049			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2050		}
2051	}
2052	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2053	spa->spa_ena = 0;
2054
2055	return (error);
2056}
2057
2058/*
2059 * Load an existing storage pool, using the pool's builtin spa_config as a
2060 * source of configuration information.
2061 */
2062static int
2063spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2064    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2065    char **ereport)
2066{
2067	int error = 0;
2068	nvlist_t *nvroot = NULL;
2069	nvlist_t *label;
2070	vdev_t *rvd;
2071	uberblock_t *ub = &spa->spa_uberblock;
2072	uint64_t children, config_cache_txg = spa->spa_config_txg;
2073	int orig_mode = spa->spa_mode;
2074	int parse;
2075	uint64_t obj;
2076	boolean_t missing_feat_write = B_FALSE;
2077
2078	/*
2079	 * If this is an untrusted config, access the pool in read-only mode.
2080	 * This prevents things like resilvering recently removed devices.
2081	 */
2082	if (!mosconfig)
2083		spa->spa_mode = FREAD;
2084
2085	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2086
2087	spa->spa_load_state = state;
2088
2089	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2090		return (SET_ERROR(EINVAL));
2091
2092	parse = (type == SPA_IMPORT_EXISTING ?
2093	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2094
2095	/*
2096	 * Create "The Godfather" zio to hold all async IOs
2097	 */
2098	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2099	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2100
2101	/*
2102	 * Parse the configuration into a vdev tree.  We explicitly set the
2103	 * value that will be returned by spa_version() since parsing the
2104	 * configuration requires knowing the version number.
2105	 */
2106	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2107	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2108	spa_config_exit(spa, SCL_ALL, FTAG);
2109
2110	if (error != 0)
2111		return (error);
2112
2113	ASSERT(spa->spa_root_vdev == rvd);
2114
2115	if (type != SPA_IMPORT_ASSEMBLE) {
2116		ASSERT(spa_guid(spa) == pool_guid);
2117	}
2118
2119	/*
2120	 * Try to open all vdevs, loading each label in the process.
2121	 */
2122	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2123	error = vdev_open(rvd);
2124	spa_config_exit(spa, SCL_ALL, FTAG);
2125	if (error != 0)
2126		return (error);
2127
2128	/*
2129	 * We need to validate the vdev labels against the configuration that
2130	 * we have in hand, which is dependent on the setting of mosconfig. If
2131	 * mosconfig is true then we're validating the vdev labels based on
2132	 * that config.  Otherwise, we're validating against the cached config
2133	 * (zpool.cache) that was read when we loaded the zfs module, and then
2134	 * later we will recursively call spa_load() and validate against
2135	 * the vdev config.
2136	 *
2137	 * If we're assembling a new pool that's been split off from an
2138	 * existing pool, the labels haven't yet been updated so we skip
2139	 * validation for now.
2140	 */
2141	if (type != SPA_IMPORT_ASSEMBLE) {
2142		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2143		error = vdev_validate(rvd, mosconfig);
2144		spa_config_exit(spa, SCL_ALL, FTAG);
2145
2146		if (error != 0)
2147			return (error);
2148
2149		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2150			return (SET_ERROR(ENXIO));
2151	}
2152
2153	/*
2154	 * Find the best uberblock.
2155	 */
2156	vdev_uberblock_load(rvd, ub, &label);
2157
2158	/*
2159	 * If we weren't able to find a single valid uberblock, return failure.
2160	 */
2161	if (ub->ub_txg == 0) {
2162		nvlist_free(label);
2163		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2164	}
2165
2166	/*
2167	 * If the pool has an unsupported version we can't open it.
2168	 */
2169	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2170		nvlist_free(label);
2171		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2172	}
2173
2174	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2175		nvlist_t *features;
2176
2177		/*
2178		 * If we weren't able to find what's necessary for reading the
2179		 * MOS in the label, return failure.
2180		 */
2181		if (label == NULL || nvlist_lookup_nvlist(label,
2182		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2183			nvlist_free(label);
2184			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2185			    ENXIO));
2186		}
2187
2188		/*
2189		 * Update our in-core representation with the definitive values
2190		 * from the label.
2191		 */
2192		nvlist_free(spa->spa_label_features);
2193		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2194	}
2195
2196	nvlist_free(label);
2197
2198	/*
2199	 * Look through entries in the label nvlist's features_for_read. If
2200	 * there is a feature listed there which we don't understand then we
2201	 * cannot open a pool.
2202	 */
2203	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2204		nvlist_t *unsup_feat;
2205
2206		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2207		    0);
2208
2209		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2210		    NULL); nvp != NULL;
2211		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2212			if (!zfeature_is_supported(nvpair_name(nvp))) {
2213				VERIFY(nvlist_add_string(unsup_feat,
2214				    nvpair_name(nvp), "") == 0);
2215			}
2216		}
2217
2218		if (!nvlist_empty(unsup_feat)) {
2219			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2220			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2221			nvlist_free(unsup_feat);
2222			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2223			    ENOTSUP));
2224		}
2225
2226		nvlist_free(unsup_feat);
2227	}
2228
2229	/*
2230	 * If the vdev guid sum doesn't match the uberblock, we have an
2231	 * incomplete configuration.  We first check to see if the pool
2232	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2233	 * If it is, defer the vdev_guid_sum check till later so we
2234	 * can handle missing vdevs.
2235	 */
2236	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2237	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2238	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2239		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2240
2241	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2242		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2243		spa_try_repair(spa, config);
2244		spa_config_exit(spa, SCL_ALL, FTAG);
2245		nvlist_free(spa->spa_config_splitting);
2246		spa->spa_config_splitting = NULL;
2247	}
2248
2249	/*
2250	 * Initialize internal SPA structures.
2251	 */
2252	spa->spa_state = POOL_STATE_ACTIVE;
2253	spa->spa_ubsync = spa->spa_uberblock;
2254	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2255	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2256	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2257	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2258	spa->spa_claim_max_txg = spa->spa_first_txg;
2259	spa->spa_prev_software_version = ub->ub_software_version;
2260
2261	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2262	if (error)
2263		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2264	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2265
2266	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2267		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2268
2269	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2270		boolean_t missing_feat_read = B_FALSE;
2271		nvlist_t *unsup_feat, *enabled_feat;
2272
2273		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2274		    &spa->spa_feat_for_read_obj) != 0) {
2275			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276		}
2277
2278		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2279		    &spa->spa_feat_for_write_obj) != 0) {
2280			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2281		}
2282
2283		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2284		    &spa->spa_feat_desc_obj) != 0) {
2285			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2286		}
2287
2288		enabled_feat = fnvlist_alloc();
2289		unsup_feat = fnvlist_alloc();
2290
2291		if (!feature_is_supported(spa->spa_meta_objset,
2292		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2293		    unsup_feat, enabled_feat))
2294			missing_feat_read = B_TRUE;
2295
2296		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2297			if (!feature_is_supported(spa->spa_meta_objset,
2298			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2299			    unsup_feat, enabled_feat)) {
2300				missing_feat_write = B_TRUE;
2301			}
2302		}
2303
2304		fnvlist_add_nvlist(spa->spa_load_info,
2305		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2306
2307		if (!nvlist_empty(unsup_feat)) {
2308			fnvlist_add_nvlist(spa->spa_load_info,
2309			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2310		}
2311
2312		fnvlist_free(enabled_feat);
2313		fnvlist_free(unsup_feat);
2314
2315		if (!missing_feat_read) {
2316			fnvlist_add_boolean(spa->spa_load_info,
2317			    ZPOOL_CONFIG_CAN_RDONLY);
2318		}
2319
2320		/*
2321		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2322		 * twofold: to determine whether the pool is available for
2323		 * import in read-write mode and (if it is not) whether the
2324		 * pool is available for import in read-only mode. If the pool
2325		 * is available for import in read-write mode, it is displayed
2326		 * as available in userland; if it is not available for import
2327		 * in read-only mode, it is displayed as unavailable in
2328		 * userland. If the pool is available for import in read-only
2329		 * mode but not read-write mode, it is displayed as unavailable
2330		 * in userland with a special note that the pool is actually
2331		 * available for open in read-only mode.
2332		 *
2333		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2334		 * missing a feature for write, we must first determine whether
2335		 * the pool can be opened read-only before returning to
2336		 * userland in order to know whether to display the
2337		 * abovementioned note.
2338		 */
2339		if (missing_feat_read || (missing_feat_write &&
2340		    spa_writeable(spa))) {
2341			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2342			    ENOTSUP));
2343		}
2344	}
2345
2346	spa->spa_is_initializing = B_TRUE;
2347	error = dsl_pool_open(spa->spa_dsl_pool);
2348	spa->spa_is_initializing = B_FALSE;
2349	if (error != 0)
2350		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2351
2352	if (!mosconfig) {
2353		uint64_t hostid;
2354		nvlist_t *policy = NULL, *nvconfig;
2355
2356		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2357			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2358
2359		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2360		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2361			char *hostname;
2362			unsigned long myhostid = 0;
2363
2364			VERIFY(nvlist_lookup_string(nvconfig,
2365			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2366
2367#ifdef	_KERNEL
2368			myhostid = zone_get_hostid(NULL);
2369#else	/* _KERNEL */
2370			/*
2371			 * We're emulating the system's hostid in userland, so
2372			 * we can't use zone_get_hostid().
2373			 */
2374			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2375#endif	/* _KERNEL */
2376			if (hostid != 0 && myhostid != 0 &&
2377			    hostid != myhostid) {
2378				nvlist_free(nvconfig);
2379				cmn_err(CE_WARN, "pool '%s' could not be "
2380				    "loaded as it was last accessed by "
2381				    "another system (host: %s hostid: 0x%lx). "
2382				    "See: http://illumos.org/msg/ZFS-8000-EY",
2383				    spa_name(spa), hostname,
2384				    (unsigned long)hostid);
2385				return (SET_ERROR(EBADF));
2386			}
2387		}
2388		if (nvlist_lookup_nvlist(spa->spa_config,
2389		    ZPOOL_REWIND_POLICY, &policy) == 0)
2390			VERIFY(nvlist_add_nvlist(nvconfig,
2391			    ZPOOL_REWIND_POLICY, policy) == 0);
2392
2393		spa_config_set(spa, nvconfig);
2394		spa_unload(spa);
2395		spa_deactivate(spa);
2396		spa_activate(spa, orig_mode);
2397
2398		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2399	}
2400
2401	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2402		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2403	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2404	if (error != 0)
2405		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2406
2407	/*
2408	 * Load the bit that tells us to use the new accounting function
2409	 * (raid-z deflation).  If we have an older pool, this will not
2410	 * be present.
2411	 */
2412	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2413	if (error != 0 && error != ENOENT)
2414		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2415
2416	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2417	    &spa->spa_creation_version);
2418	if (error != 0 && error != ENOENT)
2419		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2420
2421	/*
2422	 * Load the persistent error log.  If we have an older pool, this will
2423	 * not be present.
2424	 */
2425	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2426	if (error != 0 && error != ENOENT)
2427		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2428
2429	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2430	    &spa->spa_errlog_scrub);
2431	if (error != 0 && error != ENOENT)
2432		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2433
2434	/*
2435	 * Load the history object.  If we have an older pool, this
2436	 * will not be present.
2437	 */
2438	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2439	if (error != 0 && error != ENOENT)
2440		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2441
2442	/*
2443	 * If we're assembling the pool from the split-off vdevs of
2444	 * an existing pool, we don't want to attach the spares & cache
2445	 * devices.
2446	 */
2447
2448	/*
2449	 * Load any hot spares for this pool.
2450	 */
2451	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2452	if (error != 0 && error != ENOENT)
2453		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2454	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2455		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2456		if (load_nvlist(spa, spa->spa_spares.sav_object,
2457		    &spa->spa_spares.sav_config) != 0)
2458			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2459
2460		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2461		spa_load_spares(spa);
2462		spa_config_exit(spa, SCL_ALL, FTAG);
2463	} else if (error == 0) {
2464		spa->spa_spares.sav_sync = B_TRUE;
2465	}
2466
2467	/*
2468	 * Load any level 2 ARC devices for this pool.
2469	 */
2470	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2471	    &spa->spa_l2cache.sav_object);
2472	if (error != 0 && error != ENOENT)
2473		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2474	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2475		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2476		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2477		    &spa->spa_l2cache.sav_config) != 0)
2478			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2479
2480		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2481		spa_load_l2cache(spa);
2482		spa_config_exit(spa, SCL_ALL, FTAG);
2483	} else if (error == 0) {
2484		spa->spa_l2cache.sav_sync = B_TRUE;
2485	}
2486
2487	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2488
2489	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2490	if (error && error != ENOENT)
2491		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2492
2493	if (error == 0) {
2494		uint64_t autoreplace;
2495
2496		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2497		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2498		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2499		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2500		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2501		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2502		    &spa->spa_dedup_ditto);
2503
2504		spa->spa_autoreplace = (autoreplace != 0);
2505	}
2506
2507	/*
2508	 * If the 'autoreplace' property is set, then post a resource notifying
2509	 * the ZFS DE that it should not issue any faults for unopenable
2510	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2511	 * unopenable vdevs so that the normal autoreplace handler can take
2512	 * over.
2513	 */
2514	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2515		spa_check_removed(spa->spa_root_vdev);
2516		/*
2517		 * For the import case, this is done in spa_import(), because
2518		 * at this point we're using the spare definitions from
2519		 * the MOS config, not necessarily from the userland config.
2520		 */
2521		if (state != SPA_LOAD_IMPORT) {
2522			spa_aux_check_removed(&spa->spa_spares);
2523			spa_aux_check_removed(&spa->spa_l2cache);
2524		}
2525	}
2526
2527	/*
2528	 * Load the vdev state for all toplevel vdevs.
2529	 */
2530	vdev_load(rvd);
2531
2532	/*
2533	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2534	 */
2535	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2536	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2537	spa_config_exit(spa, SCL_ALL, FTAG);
2538
2539	/*
2540	 * Load the DDTs (dedup tables).
2541	 */
2542	error = ddt_load(spa);
2543	if (error != 0)
2544		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2545
2546	spa_update_dspace(spa);
2547
2548	/*
2549	 * Validate the config, using the MOS config to fill in any
2550	 * information which might be missing.  If we fail to validate
2551	 * the config then declare the pool unfit for use. If we're
2552	 * assembling a pool from a split, the log is not transferred
2553	 * over.
2554	 */
2555	if (type != SPA_IMPORT_ASSEMBLE) {
2556		nvlist_t *nvconfig;
2557
2558		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2559			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2560
2561		if (!spa_config_valid(spa, nvconfig)) {
2562			nvlist_free(nvconfig);
2563			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2564			    ENXIO));
2565		}
2566		nvlist_free(nvconfig);
2567
2568		/*
2569		 * Now that we've validated the config, check the state of the
2570		 * root vdev.  If it can't be opened, it indicates one or
2571		 * more toplevel vdevs are faulted.
2572		 */
2573		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2574			return (SET_ERROR(ENXIO));
2575
2576		if (spa_check_logs(spa)) {
2577			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2578			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2579		}
2580	}
2581
2582	if (missing_feat_write) {
2583		ASSERT(state == SPA_LOAD_TRYIMPORT);
2584
2585		/*
2586		 * At this point, we know that we can open the pool in
2587		 * read-only mode but not read-write mode. We now have enough
2588		 * information and can return to userland.
2589		 */
2590		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2591	}
2592
2593	/*
2594	 * We've successfully opened the pool, verify that we're ready
2595	 * to start pushing transactions.
2596	 */
2597	if (state != SPA_LOAD_TRYIMPORT) {
2598		if (error = spa_load_verify(spa))
2599			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2600			    error));
2601	}
2602
2603	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2604	    spa->spa_load_max_txg == UINT64_MAX)) {
2605		dmu_tx_t *tx;
2606		int need_update = B_FALSE;
2607
2608		ASSERT(state != SPA_LOAD_TRYIMPORT);
2609
2610		/*
2611		 * Claim log blocks that haven't been committed yet.
2612		 * This must all happen in a single txg.
2613		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2614		 * invoked from zil_claim_log_block()'s i/o done callback.
2615		 * Price of rollback is that we abandon the log.
2616		 */
2617		spa->spa_claiming = B_TRUE;
2618
2619		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2620		    spa_first_txg(spa));
2621		(void) dmu_objset_find(spa_name(spa),
2622		    zil_claim, tx, DS_FIND_CHILDREN);
2623		dmu_tx_commit(tx);
2624
2625		spa->spa_claiming = B_FALSE;
2626
2627		spa_set_log_state(spa, SPA_LOG_GOOD);
2628		spa->spa_sync_on = B_TRUE;
2629		txg_sync_start(spa->spa_dsl_pool);
2630
2631		/*
2632		 * Wait for all claims to sync.  We sync up to the highest
2633		 * claimed log block birth time so that claimed log blocks
2634		 * don't appear to be from the future.  spa_claim_max_txg
2635		 * will have been set for us by either zil_check_log_chain()
2636		 * (invoked from spa_check_logs()) or zil_claim() above.
2637		 */
2638		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2639
2640		/*
2641		 * If the config cache is stale, or we have uninitialized
2642		 * metaslabs (see spa_vdev_add()), then update the config.
2643		 *
2644		 * If this is a verbatim import, trust the current
2645		 * in-core spa_config and update the disk labels.
2646		 */
2647		if (config_cache_txg != spa->spa_config_txg ||
2648		    state == SPA_LOAD_IMPORT ||
2649		    state == SPA_LOAD_RECOVER ||
2650		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2651			need_update = B_TRUE;
2652
2653		for (int c = 0; c < rvd->vdev_children; c++)
2654			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2655				need_update = B_TRUE;
2656
2657		/*
2658		 * Update the config cache asychronously in case we're the
2659		 * root pool, in which case the config cache isn't writable yet.
2660		 */
2661		if (need_update)
2662			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2663
2664		/*
2665		 * Check all DTLs to see if anything needs resilvering.
2666		 */
2667		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2668		    vdev_resilver_needed(rvd, NULL, NULL))
2669			spa_async_request(spa, SPA_ASYNC_RESILVER);
2670
2671		/*
2672		 * Log the fact that we booted up (so that we can detect if
2673		 * we rebooted in the middle of an operation).
2674		 */
2675		spa_history_log_version(spa, "open");
2676
2677		/*
2678		 * Delete any inconsistent datasets.
2679		 */
2680		(void) dmu_objset_find(spa_name(spa),
2681		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2682
2683		/*
2684		 * Clean up any stale temporary dataset userrefs.
2685		 */
2686		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2687	}
2688
2689	return (0);
2690}
2691
2692static int
2693spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2694{
2695	int mode = spa->spa_mode;
2696
2697	spa_unload(spa);
2698	spa_deactivate(spa);
2699
2700	spa->spa_load_max_txg--;
2701
2702	spa_activate(spa, mode);
2703	spa_async_suspend(spa);
2704
2705	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2706}
2707
2708/*
2709 * If spa_load() fails this function will try loading prior txg's. If
2710 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2711 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2712 * function will not rewind the pool and will return the same error as
2713 * spa_load().
2714 */
2715static int
2716spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2717    uint64_t max_request, int rewind_flags)
2718{
2719	nvlist_t *loadinfo = NULL;
2720	nvlist_t *config = NULL;
2721	int load_error, rewind_error;
2722	uint64_t safe_rewind_txg;
2723	uint64_t min_txg;
2724
2725	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2726		spa->spa_load_max_txg = spa->spa_load_txg;
2727		spa_set_log_state(spa, SPA_LOG_CLEAR);
2728	} else {
2729		spa->spa_load_max_txg = max_request;
2730	}
2731
2732	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2733	    mosconfig);
2734	if (load_error == 0)
2735		return (0);
2736
2737	if (spa->spa_root_vdev != NULL)
2738		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2739
2740	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2741	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2742
2743	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2744		nvlist_free(config);
2745		return (load_error);
2746	}
2747
2748	if (state == SPA_LOAD_RECOVER) {
2749		/* Price of rolling back is discarding txgs, including log */
2750		spa_set_log_state(spa, SPA_LOG_CLEAR);
2751	} else {
2752		/*
2753		 * If we aren't rolling back save the load info from our first
2754		 * import attempt so that we can restore it after attempting
2755		 * to rewind.
2756		 */
2757		loadinfo = spa->spa_load_info;
2758		spa->spa_load_info = fnvlist_alloc();
2759	}
2760
2761	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2762	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2763	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2764	    TXG_INITIAL : safe_rewind_txg;
2765
2766	/*
2767	 * Continue as long as we're finding errors, we're still within
2768	 * the acceptable rewind range, and we're still finding uberblocks
2769	 */
2770	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2771	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2772		if (spa->spa_load_max_txg < safe_rewind_txg)
2773			spa->spa_extreme_rewind = B_TRUE;
2774		rewind_error = spa_load_retry(spa, state, mosconfig);
2775	}
2776
2777	spa->spa_extreme_rewind = B_FALSE;
2778	spa->spa_load_max_txg = UINT64_MAX;
2779
2780	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2781		spa_config_set(spa, config);
2782
2783	if (state == SPA_LOAD_RECOVER) {
2784		ASSERT3P(loadinfo, ==, NULL);
2785		return (rewind_error);
2786	} else {
2787		/* Store the rewind info as part of the initial load info */
2788		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2789		    spa->spa_load_info);
2790
2791		/* Restore the initial load info */
2792		fnvlist_free(spa->spa_load_info);
2793		spa->spa_load_info = loadinfo;
2794
2795		return (load_error);
2796	}
2797}
2798
2799/*
2800 * Pool Open/Import
2801 *
2802 * The import case is identical to an open except that the configuration is sent
2803 * down from userland, instead of grabbed from the configuration cache.  For the
2804 * case of an open, the pool configuration will exist in the
2805 * POOL_STATE_UNINITIALIZED state.
2806 *
2807 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2808 * the same time open the pool, without having to keep around the spa_t in some
2809 * ambiguous state.
2810 */
2811static int
2812spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2813    nvlist_t **config)
2814{
2815	spa_t *spa;
2816	spa_load_state_t state = SPA_LOAD_OPEN;
2817	int error;
2818	int locked = B_FALSE;
2819
2820	*spapp = NULL;
2821
2822	/*
2823	 * As disgusting as this is, we need to support recursive calls to this
2824	 * function because dsl_dir_open() is called during spa_load(), and ends
2825	 * up calling spa_open() again.  The real fix is to figure out how to
2826	 * avoid dsl_dir_open() calling this in the first place.
2827	 */
2828	if (mutex_owner(&spa_namespace_lock) != curthread) {
2829		mutex_enter(&spa_namespace_lock);
2830		locked = B_TRUE;
2831	}
2832
2833	if ((spa = spa_lookup(pool)) == NULL) {
2834		if (locked)
2835			mutex_exit(&spa_namespace_lock);
2836		return (SET_ERROR(ENOENT));
2837	}
2838
2839	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2840		zpool_rewind_policy_t policy;
2841
2842		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2843		    &policy);
2844		if (policy.zrp_request & ZPOOL_DO_REWIND)
2845			state = SPA_LOAD_RECOVER;
2846
2847		spa_activate(spa, spa_mode_global);
2848
2849		if (state != SPA_LOAD_RECOVER)
2850			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2851
2852		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2853		    policy.zrp_request);
2854
2855		if (error == EBADF) {
2856			/*
2857			 * If vdev_validate() returns failure (indicated by
2858			 * EBADF), it indicates that one of the vdevs indicates
2859			 * that the pool has been exported or destroyed.  If
2860			 * this is the case, the config cache is out of sync and
2861			 * we should remove the pool from the namespace.
2862			 */
2863			spa_unload(spa);
2864			spa_deactivate(spa);
2865			spa_config_sync(spa, B_TRUE, B_TRUE);
2866			spa_remove(spa);
2867			if (locked)
2868				mutex_exit(&spa_namespace_lock);
2869			return (SET_ERROR(ENOENT));
2870		}
2871
2872		if (error) {
2873			/*
2874			 * We can't open the pool, but we still have useful
2875			 * information: the state of each vdev after the
2876			 * attempted vdev_open().  Return this to the user.
2877			 */
2878			if (config != NULL && spa->spa_config) {
2879				VERIFY(nvlist_dup(spa->spa_config, config,
2880				    KM_SLEEP) == 0);
2881				VERIFY(nvlist_add_nvlist(*config,
2882				    ZPOOL_CONFIG_LOAD_INFO,
2883				    spa->spa_load_info) == 0);
2884			}
2885			spa_unload(spa);
2886			spa_deactivate(spa);
2887			spa->spa_last_open_failed = error;
2888			if (locked)
2889				mutex_exit(&spa_namespace_lock);
2890			*spapp = NULL;
2891			return (error);
2892		}
2893	}
2894
2895	spa_open_ref(spa, tag);
2896
2897	if (config != NULL)
2898		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2899
2900	/*
2901	 * If we've recovered the pool, pass back any information we
2902	 * gathered while doing the load.
2903	 */
2904	if (state == SPA_LOAD_RECOVER) {
2905		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2906		    spa->spa_load_info) == 0);
2907	}
2908
2909	if (locked) {
2910		spa->spa_last_open_failed = 0;
2911		spa->spa_last_ubsync_txg = 0;
2912		spa->spa_load_txg = 0;
2913		mutex_exit(&spa_namespace_lock);
2914	}
2915
2916	*spapp = spa;
2917
2918	return (0);
2919}
2920
2921int
2922spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2923    nvlist_t **config)
2924{
2925	return (spa_open_common(name, spapp, tag, policy, config));
2926}
2927
2928int
2929spa_open(const char *name, spa_t **spapp, void *tag)
2930{
2931	return (spa_open_common(name, spapp, tag, NULL, NULL));
2932}
2933
2934/*
2935 * Lookup the given spa_t, incrementing the inject count in the process,
2936 * preventing it from being exported or destroyed.
2937 */
2938spa_t *
2939spa_inject_addref(char *name)
2940{
2941	spa_t *spa;
2942
2943	mutex_enter(&spa_namespace_lock);
2944	if ((spa = spa_lookup(name)) == NULL) {
2945		mutex_exit(&spa_namespace_lock);
2946		return (NULL);
2947	}
2948	spa->spa_inject_ref++;
2949	mutex_exit(&spa_namespace_lock);
2950
2951	return (spa);
2952}
2953
2954void
2955spa_inject_delref(spa_t *spa)
2956{
2957	mutex_enter(&spa_namespace_lock);
2958	spa->spa_inject_ref--;
2959	mutex_exit(&spa_namespace_lock);
2960}
2961
2962/*
2963 * Add spares device information to the nvlist.
2964 */
2965static void
2966spa_add_spares(spa_t *spa, nvlist_t *config)
2967{
2968	nvlist_t **spares;
2969	uint_t i, nspares;
2970	nvlist_t *nvroot;
2971	uint64_t guid;
2972	vdev_stat_t *vs;
2973	uint_t vsc;
2974	uint64_t pool;
2975
2976	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2977
2978	if (spa->spa_spares.sav_count == 0)
2979		return;
2980
2981	VERIFY(nvlist_lookup_nvlist(config,
2982	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2983	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2984	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2985	if (nspares != 0) {
2986		VERIFY(nvlist_add_nvlist_array(nvroot,
2987		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2988		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2989		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2990
2991		/*
2992		 * Go through and find any spares which have since been
2993		 * repurposed as an active spare.  If this is the case, update
2994		 * their status appropriately.
2995		 */
2996		for (i = 0; i < nspares; i++) {
2997			VERIFY(nvlist_lookup_uint64(spares[i],
2998			    ZPOOL_CONFIG_GUID, &guid) == 0);
2999			if (spa_spare_exists(guid, &pool, NULL) &&
3000			    pool != 0ULL) {
3001				VERIFY(nvlist_lookup_uint64_array(
3002				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3003				    (uint64_t **)&vs, &vsc) == 0);
3004				vs->vs_state = VDEV_STATE_CANT_OPEN;
3005				vs->vs_aux = VDEV_AUX_SPARED;
3006			}
3007		}
3008	}
3009}
3010
3011/*
3012 * Add l2cache device information to the nvlist, including vdev stats.
3013 */
3014static void
3015spa_add_l2cache(spa_t *spa, nvlist_t *config)
3016{
3017	nvlist_t **l2cache;
3018	uint_t i, j, nl2cache;
3019	nvlist_t *nvroot;
3020	uint64_t guid;
3021	vdev_t *vd;
3022	vdev_stat_t *vs;
3023	uint_t vsc;
3024
3025	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3026
3027	if (spa->spa_l2cache.sav_count == 0)
3028		return;
3029
3030	VERIFY(nvlist_lookup_nvlist(config,
3031	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3032	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3033	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3034	if (nl2cache != 0) {
3035		VERIFY(nvlist_add_nvlist_array(nvroot,
3036		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3037		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3038		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3039
3040		/*
3041		 * Update level 2 cache device stats.
3042		 */
3043
3044		for (i = 0; i < nl2cache; i++) {
3045			VERIFY(nvlist_lookup_uint64(l2cache[i],
3046			    ZPOOL_CONFIG_GUID, &guid) == 0);
3047
3048			vd = NULL;
3049			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3050				if (guid ==
3051				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3052					vd = spa->spa_l2cache.sav_vdevs[j];
3053					break;
3054				}
3055			}
3056			ASSERT(vd != NULL);
3057
3058			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3059			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3060			    == 0);
3061			vdev_get_stats(vd, vs);
3062		}
3063	}
3064}
3065
3066static void
3067spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3068{
3069	nvlist_t *features;
3070	zap_cursor_t zc;
3071	zap_attribute_t za;
3072
3073	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3074	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3075
3076	if (spa->spa_feat_for_read_obj != 0) {
3077		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3078		    spa->spa_feat_for_read_obj);
3079		    zap_cursor_retrieve(&zc, &za) == 0;
3080		    zap_cursor_advance(&zc)) {
3081			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3082			    za.za_num_integers == 1);
3083			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3084			    za.za_first_integer));
3085		}
3086		zap_cursor_fini(&zc);
3087	}
3088
3089	if (spa->spa_feat_for_write_obj != 0) {
3090		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3091		    spa->spa_feat_for_write_obj);
3092		    zap_cursor_retrieve(&zc, &za) == 0;
3093		    zap_cursor_advance(&zc)) {
3094			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3095			    za.za_num_integers == 1);
3096			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3097			    za.za_first_integer));
3098		}
3099		zap_cursor_fini(&zc);
3100	}
3101
3102	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3103	    features) == 0);
3104	nvlist_free(features);
3105}
3106
3107int
3108spa_get_stats(const char *name, nvlist_t **config,
3109    char *altroot, size_t buflen)
3110{
3111	int error;
3112	spa_t *spa;
3113
3114	*config = NULL;
3115	error = spa_open_common(name, &spa, FTAG, NULL, config);
3116
3117	if (spa != NULL) {
3118		/*
3119		 * This still leaves a window of inconsistency where the spares
3120		 * or l2cache devices could change and the config would be
3121		 * self-inconsistent.
3122		 */
3123		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3124
3125		if (*config != NULL) {
3126			uint64_t loadtimes[2];
3127
3128			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3129			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3130			VERIFY(nvlist_add_uint64_array(*config,
3131			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3132
3133			VERIFY(nvlist_add_uint64(*config,
3134			    ZPOOL_CONFIG_ERRCOUNT,
3135			    spa_get_errlog_size(spa)) == 0);
3136
3137			if (spa_suspended(spa))
3138				VERIFY(nvlist_add_uint64(*config,
3139				    ZPOOL_CONFIG_SUSPENDED,
3140				    spa->spa_failmode) == 0);
3141
3142			spa_add_spares(spa, *config);
3143			spa_add_l2cache(spa, *config);
3144			spa_add_feature_stats(spa, *config);
3145		}
3146	}
3147
3148	/*
3149	 * We want to get the alternate root even for faulted pools, so we cheat
3150	 * and call spa_lookup() directly.
3151	 */
3152	if (altroot) {
3153		if (spa == NULL) {
3154			mutex_enter(&spa_namespace_lock);
3155			spa = spa_lookup(name);
3156			if (spa)
3157				spa_altroot(spa, altroot, buflen);
3158			else
3159				altroot[0] = '\0';
3160			spa = NULL;
3161			mutex_exit(&spa_namespace_lock);
3162		} else {
3163			spa_altroot(spa, altroot, buflen);
3164		}
3165	}
3166
3167	if (spa != NULL) {
3168		spa_config_exit(spa, SCL_CONFIG, FTAG);
3169		spa_close(spa, FTAG);
3170	}
3171
3172	return (error);
3173}
3174
3175/*
3176 * Validate that the auxiliary device array is well formed.  We must have an
3177 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3178 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3179 * specified, as long as they are well-formed.
3180 */
3181static int
3182spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3183    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3184    vdev_labeltype_t label)
3185{
3186	nvlist_t **dev;
3187	uint_t i, ndev;
3188	vdev_t *vd;
3189	int error;
3190
3191	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3192
3193	/*
3194	 * It's acceptable to have no devs specified.
3195	 */
3196	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3197		return (0);
3198
3199	if (ndev == 0)
3200		return (SET_ERROR(EINVAL));
3201
3202	/*
3203	 * Make sure the pool is formatted with a version that supports this
3204	 * device type.
3205	 */
3206	if (spa_version(spa) < version)
3207		return (SET_ERROR(ENOTSUP));
3208
3209	/*
3210	 * Set the pending device list so we correctly handle device in-use
3211	 * checking.
3212	 */
3213	sav->sav_pending = dev;
3214	sav->sav_npending = ndev;
3215
3216	for (i = 0; i < ndev; i++) {
3217		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3218		    mode)) != 0)
3219			goto out;
3220
3221		if (!vd->vdev_ops->vdev_op_leaf) {
3222			vdev_free(vd);
3223			error = SET_ERROR(EINVAL);
3224			goto out;
3225		}
3226
3227		/*
3228		 * The L2ARC currently only supports disk devices in
3229		 * kernel context.  For user-level testing, we allow it.
3230		 */
3231#ifdef _KERNEL
3232		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3233		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3234			error = SET_ERROR(ENOTBLK);
3235			vdev_free(vd);
3236			goto out;
3237		}
3238#endif
3239		vd->vdev_top = vd;
3240
3241		if ((error = vdev_open(vd)) == 0 &&
3242		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3243			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3244			    vd->vdev_guid) == 0);
3245		}
3246
3247		vdev_free(vd);
3248
3249		if (error &&
3250		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3251			goto out;
3252		else
3253			error = 0;
3254	}
3255
3256out:
3257	sav->sav_pending = NULL;
3258	sav->sav_npending = 0;
3259	return (error);
3260}
3261
3262static int
3263spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3264{
3265	int error;
3266
3267	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3268
3269	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3270	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3271	    VDEV_LABEL_SPARE)) != 0) {
3272		return (error);
3273	}
3274
3275	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3276	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3277	    VDEV_LABEL_L2CACHE));
3278}
3279
3280static void
3281spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3282    const char *config)
3283{
3284	int i;
3285
3286	if (sav->sav_config != NULL) {
3287		nvlist_t **olddevs;
3288		uint_t oldndevs;
3289		nvlist_t **newdevs;
3290
3291		/*
3292		 * Generate new dev list by concatentating with the
3293		 * current dev list.
3294		 */
3295		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3296		    &olddevs, &oldndevs) == 0);
3297
3298		newdevs = kmem_alloc(sizeof (void *) *
3299		    (ndevs + oldndevs), KM_SLEEP);
3300		for (i = 0; i < oldndevs; i++)
3301			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3302			    KM_SLEEP) == 0);
3303		for (i = 0; i < ndevs; i++)
3304			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3305			    KM_SLEEP) == 0);
3306
3307		VERIFY(nvlist_remove(sav->sav_config, config,
3308		    DATA_TYPE_NVLIST_ARRAY) == 0);
3309
3310		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3311		    config, newdevs, ndevs + oldndevs) == 0);
3312		for (i = 0; i < oldndevs + ndevs; i++)
3313			nvlist_free(newdevs[i]);
3314		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3315	} else {
3316		/*
3317		 * Generate a new dev list.
3318		 */
3319		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3320		    KM_SLEEP) == 0);
3321		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3322		    devs, ndevs) == 0);
3323	}
3324}
3325
3326/*
3327 * Stop and drop level 2 ARC devices
3328 */
3329void
3330spa_l2cache_drop(spa_t *spa)
3331{
3332	vdev_t *vd;
3333	int i;
3334	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3335
3336	for (i = 0; i < sav->sav_count; i++) {
3337		uint64_t pool;
3338
3339		vd = sav->sav_vdevs[i];
3340		ASSERT(vd != NULL);
3341
3342		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3343		    pool != 0ULL && l2arc_vdev_present(vd))
3344			l2arc_remove_vdev(vd);
3345	}
3346}
3347
3348/*
3349 * Pool Creation
3350 */
3351int
3352spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3353    nvlist_t *zplprops)
3354{
3355	spa_t *spa;
3356	char *altroot = NULL;
3357	vdev_t *rvd;
3358	dsl_pool_t *dp;
3359	dmu_tx_t *tx;
3360	int error = 0;
3361	uint64_t txg = TXG_INITIAL;
3362	nvlist_t **spares, **l2cache;
3363	uint_t nspares, nl2cache;
3364	uint64_t version, obj;
3365	boolean_t has_features;
3366
3367	/*
3368	 * If this pool already exists, return failure.
3369	 */
3370	mutex_enter(&spa_namespace_lock);
3371	if (spa_lookup(pool) != NULL) {
3372		mutex_exit(&spa_namespace_lock);
3373		return (SET_ERROR(EEXIST));
3374	}
3375
3376	/*
3377	 * Allocate a new spa_t structure.
3378	 */
3379	(void) nvlist_lookup_string(props,
3380	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3381	spa = spa_add(pool, NULL, altroot);
3382	spa_activate(spa, spa_mode_global);
3383
3384	if (props && (error = spa_prop_validate(spa, props))) {
3385		spa_deactivate(spa);
3386		spa_remove(spa);
3387		mutex_exit(&spa_namespace_lock);
3388		return (error);
3389	}
3390
3391	has_features = B_FALSE;
3392	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3393	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3394		if (zpool_prop_feature(nvpair_name(elem)))
3395			has_features = B_TRUE;
3396	}
3397
3398	if (has_features || nvlist_lookup_uint64(props,
3399	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3400		version = SPA_VERSION;
3401	}
3402	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3403
3404	spa->spa_first_txg = txg;
3405	spa->spa_uberblock.ub_txg = txg - 1;
3406	spa->spa_uberblock.ub_version = version;
3407	spa->spa_ubsync = spa->spa_uberblock;
3408
3409	/*
3410	 * Create "The Godfather" zio to hold all async IOs
3411	 */
3412	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3413	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3414
3415	/*
3416	 * Create the root vdev.
3417	 */
3418	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3419
3420	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3421
3422	ASSERT(error != 0 || rvd != NULL);
3423	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3424
3425	if (error == 0 && !zfs_allocatable_devs(nvroot))
3426		error = SET_ERROR(EINVAL);
3427
3428	if (error == 0 &&
3429	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3430	    (error = spa_validate_aux(spa, nvroot, txg,
3431	    VDEV_ALLOC_ADD)) == 0) {
3432		for (int c = 0; c < rvd->vdev_children; c++) {
3433			vdev_metaslab_set_size(rvd->vdev_child[c]);
3434			vdev_expand(rvd->vdev_child[c], txg);
3435		}
3436	}
3437
3438	spa_config_exit(spa, SCL_ALL, FTAG);
3439
3440	if (error != 0) {
3441		spa_unload(spa);
3442		spa_deactivate(spa);
3443		spa_remove(spa);
3444		mutex_exit(&spa_namespace_lock);
3445		return (error);
3446	}
3447
3448	/*
3449	 * Get the list of spares, if specified.
3450	 */
3451	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3452	    &spares, &nspares) == 0) {
3453		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3454		    KM_SLEEP) == 0);
3455		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3456		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3457		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3458		spa_load_spares(spa);
3459		spa_config_exit(spa, SCL_ALL, FTAG);
3460		spa->spa_spares.sav_sync = B_TRUE;
3461	}
3462
3463	/*
3464	 * Get the list of level 2 cache devices, if specified.
3465	 */
3466	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3467	    &l2cache, &nl2cache) == 0) {
3468		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3469		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3470		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3471		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3472		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3473		spa_load_l2cache(spa);
3474		spa_config_exit(spa, SCL_ALL, FTAG);
3475		spa->spa_l2cache.sav_sync = B_TRUE;
3476	}
3477
3478	spa->spa_is_initializing = B_TRUE;
3479	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3480	spa->spa_meta_objset = dp->dp_meta_objset;
3481	spa->spa_is_initializing = B_FALSE;
3482
3483	/*
3484	 * Create DDTs (dedup tables).
3485	 */
3486	ddt_create(spa);
3487
3488	spa_update_dspace(spa);
3489
3490	tx = dmu_tx_create_assigned(dp, txg);
3491
3492	/*
3493	 * Create the pool config object.
3494	 */
3495	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3496	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3497	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3498
3499	if (zap_add(spa->spa_meta_objset,
3500	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3501	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3502		cmn_err(CE_PANIC, "failed to add pool config");
3503	}
3504
3505	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3506		spa_feature_create_zap_objects(spa, tx);
3507
3508	if (zap_add(spa->spa_meta_objset,
3509	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3510	    sizeof (uint64_t), 1, &version, tx) != 0) {
3511		cmn_err(CE_PANIC, "failed to add pool version");
3512	}
3513
3514	/* Newly created pools with the right version are always deflated. */
3515	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3516		spa->spa_deflate = TRUE;
3517		if (zap_add(spa->spa_meta_objset,
3518		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3519		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3520			cmn_err(CE_PANIC, "failed to add deflate");
3521		}
3522	}
3523
3524	/*
3525	 * Create the deferred-free bpobj.  Turn off compression
3526	 * because sync-to-convergence takes longer if the blocksize
3527	 * keeps changing.
3528	 */
3529	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3530	dmu_object_set_compress(spa->spa_meta_objset, obj,
3531	    ZIO_COMPRESS_OFF, tx);
3532	if (zap_add(spa->spa_meta_objset,
3533	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3534	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3535		cmn_err(CE_PANIC, "failed to add bpobj");
3536	}
3537	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3538	    spa->spa_meta_objset, obj));
3539
3540	/*
3541	 * Create the pool's history object.
3542	 */
3543	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3544		spa_history_create_obj(spa, tx);
3545
3546	/*
3547	 * Set pool properties.
3548	 */
3549	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3550	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3551	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3552	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3553
3554	if (props != NULL) {
3555		spa_configfile_set(spa, props, B_FALSE);
3556		spa_sync_props(props, tx);
3557	}
3558
3559	dmu_tx_commit(tx);
3560
3561	spa->spa_sync_on = B_TRUE;
3562	txg_sync_start(spa->spa_dsl_pool);
3563
3564	/*
3565	 * We explicitly wait for the first transaction to complete so that our
3566	 * bean counters are appropriately updated.
3567	 */
3568	txg_wait_synced(spa->spa_dsl_pool, txg);
3569
3570	spa_config_sync(spa, B_FALSE, B_TRUE);
3571
3572	spa_history_log_version(spa, "create");
3573
3574	spa->spa_minref = refcount_count(&spa->spa_refcount);
3575
3576	mutex_exit(&spa_namespace_lock);
3577
3578	return (0);
3579}
3580
3581#ifdef _KERNEL
3582/*
3583 * Get the root pool information from the root disk, then import the root pool
3584 * during the system boot up time.
3585 */
3586extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3587
3588static nvlist_t *
3589spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3590{
3591	nvlist_t *config;
3592	nvlist_t *nvtop, *nvroot;
3593	uint64_t pgid;
3594
3595	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3596		return (NULL);
3597
3598	/*
3599	 * Add this top-level vdev to the child array.
3600	 */
3601	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3602	    &nvtop) == 0);
3603	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3604	    &pgid) == 0);
3605	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3606
3607	/*
3608	 * Put this pool's top-level vdevs into a root vdev.
3609	 */
3610	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3611	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3612	    VDEV_TYPE_ROOT) == 0);
3613	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3614	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3615	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3616	    &nvtop, 1) == 0);
3617
3618	/*
3619	 * Replace the existing vdev_tree with the new root vdev in
3620	 * this pool's configuration (remove the old, add the new).
3621	 */
3622	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3623	nvlist_free(nvroot);
3624	return (config);
3625}
3626
3627/*
3628 * Walk the vdev tree and see if we can find a device with "better"
3629 * configuration. A configuration is "better" if the label on that
3630 * device has a more recent txg.
3631 */
3632static void
3633spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3634{
3635	for (int c = 0; c < vd->vdev_children; c++)
3636		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3637
3638	if (vd->vdev_ops->vdev_op_leaf) {
3639		nvlist_t *label;
3640		uint64_t label_txg;
3641
3642		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3643		    &label) != 0)
3644			return;
3645
3646		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3647		    &label_txg) == 0);
3648
3649		/*
3650		 * Do we have a better boot device?
3651		 */
3652		if (label_txg > *txg) {
3653			*txg = label_txg;
3654			*avd = vd;
3655		}
3656		nvlist_free(label);
3657	}
3658}
3659
3660/*
3661 * Import a root pool.
3662 *
3663 * For x86. devpath_list will consist of devid and/or physpath name of
3664 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3665 * The GRUB "findroot" command will return the vdev we should boot.
3666 *
3667 * For Sparc, devpath_list consists the physpath name of the booting device
3668 * no matter the rootpool is a single device pool or a mirrored pool.
3669 * e.g.
3670 *	"/pci@1f,0/ide@d/disk@0,0:a"
3671 */
3672int
3673spa_import_rootpool(char *devpath, char *devid)
3674{
3675	spa_t *spa;
3676	vdev_t *rvd, *bvd, *avd = NULL;
3677	nvlist_t *config, *nvtop;
3678	uint64_t guid, txg;
3679	char *pname;
3680	int error;
3681
3682	/*
3683	 * Read the label from the boot device and generate a configuration.
3684	 */
3685	config = spa_generate_rootconf(devpath, devid, &guid);
3686#if defined(_OBP) && defined(_KERNEL)
3687	if (config == NULL) {
3688		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3689			/* iscsi boot */
3690			get_iscsi_bootpath_phy(devpath);
3691			config = spa_generate_rootconf(devpath, devid, &guid);
3692		}
3693	}
3694#endif
3695	if (config == NULL) {
3696		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3697		    devpath);
3698		return (SET_ERROR(EIO));
3699	}
3700
3701	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3702	    &pname) == 0);
3703	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3704
3705	mutex_enter(&spa_namespace_lock);
3706	if ((spa = spa_lookup(pname)) != NULL) {
3707		/*
3708		 * Remove the existing root pool from the namespace so that we
3709		 * can replace it with the correct config we just read in.
3710		 */
3711		spa_remove(spa);
3712	}
3713
3714	spa = spa_add(pname, config, NULL);
3715	spa->spa_is_root = B_TRUE;
3716	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3717
3718	/*
3719	 * Build up a vdev tree based on the boot device's label config.
3720	 */
3721	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3722	    &nvtop) == 0);
3723	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3724	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3725	    VDEV_ALLOC_ROOTPOOL);
3726	spa_config_exit(spa, SCL_ALL, FTAG);
3727	if (error) {
3728		mutex_exit(&spa_namespace_lock);
3729		nvlist_free(config);
3730		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3731		    pname);
3732		return (error);
3733	}
3734
3735	/*
3736	 * Get the boot vdev.
3737	 */
3738	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3739		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3740		    (u_longlong_t)guid);
3741		error = SET_ERROR(ENOENT);
3742		goto out;
3743	}
3744
3745	/*
3746	 * Determine if there is a better boot device.
3747	 */
3748	avd = bvd;
3749	spa_alt_rootvdev(rvd, &avd, &txg);
3750	if (avd != bvd) {
3751		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3752		    "try booting from '%s'", avd->vdev_path);
3753		error = SET_ERROR(EINVAL);
3754		goto out;
3755	}
3756
3757	/*
3758	 * If the boot device is part of a spare vdev then ensure that
3759	 * we're booting off the active spare.
3760	 */
3761	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3762	    !bvd->vdev_isspare) {
3763		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3764		    "try booting from '%s'",
3765		    bvd->vdev_parent->
3766		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3767		error = SET_ERROR(EINVAL);
3768		goto out;
3769	}
3770
3771	error = 0;
3772out:
3773	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3774	vdev_free(rvd);
3775	spa_config_exit(spa, SCL_ALL, FTAG);
3776	mutex_exit(&spa_namespace_lock);
3777
3778	nvlist_free(config);
3779	return (error);
3780}
3781
3782#endif
3783
3784/*
3785 * Import a non-root pool into the system.
3786 */
3787int
3788spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3789{
3790	spa_t *spa;
3791	char *altroot = NULL;
3792	spa_load_state_t state = SPA_LOAD_IMPORT;
3793	zpool_rewind_policy_t policy;
3794	uint64_t mode = spa_mode_global;
3795	uint64_t readonly = B_FALSE;
3796	int error;
3797	nvlist_t *nvroot;
3798	nvlist_t **spares, **l2cache;
3799	uint_t nspares, nl2cache;
3800
3801	/*
3802	 * If a pool with this name exists, return failure.
3803	 */
3804	mutex_enter(&spa_namespace_lock);
3805	if (spa_lookup(pool) != NULL) {
3806		mutex_exit(&spa_namespace_lock);
3807		return (SET_ERROR(EEXIST));
3808	}
3809
3810	/*
3811	 * Create and initialize the spa structure.
3812	 */
3813	(void) nvlist_lookup_string(props,
3814	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3815	(void) nvlist_lookup_uint64(props,
3816	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3817	if (readonly)
3818		mode = FREAD;
3819	spa = spa_add(pool, config, altroot);
3820	spa->spa_import_flags = flags;
3821
3822	/*
3823	 * Verbatim import - Take a pool and insert it into the namespace
3824	 * as if it had been loaded at boot.
3825	 */
3826	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3827		if (props != NULL)
3828			spa_configfile_set(spa, props, B_FALSE);
3829
3830		spa_config_sync(spa, B_FALSE, B_TRUE);
3831
3832		mutex_exit(&spa_namespace_lock);
3833		spa_history_log_version(spa, "import");
3834
3835		return (0);
3836	}
3837
3838	spa_activate(spa, mode);
3839
3840	/*
3841	 * Don't start async tasks until we know everything is healthy.
3842	 */
3843	spa_async_suspend(spa);
3844
3845	zpool_get_rewind_policy(config, &policy);
3846	if (policy.zrp_request & ZPOOL_DO_REWIND)
3847		state = SPA_LOAD_RECOVER;
3848
3849	/*
3850	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3851	 * because the user-supplied config is actually the one to trust when
3852	 * doing an import.
3853	 */
3854	if (state != SPA_LOAD_RECOVER)
3855		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3856
3857	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3858	    policy.zrp_request);
3859
3860	/*
3861	 * Propagate anything learned while loading the pool and pass it
3862	 * back to caller (i.e. rewind info, missing devices, etc).
3863	 */
3864	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3865	    spa->spa_load_info) == 0);
3866
3867	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3868	/*
3869	 * Toss any existing sparelist, as it doesn't have any validity
3870	 * anymore, and conflicts with spa_has_spare().
3871	 */
3872	if (spa->spa_spares.sav_config) {
3873		nvlist_free(spa->spa_spares.sav_config);
3874		spa->spa_spares.sav_config = NULL;
3875		spa_load_spares(spa);
3876	}
3877	if (spa->spa_l2cache.sav_config) {
3878		nvlist_free(spa->spa_l2cache.sav_config);
3879		spa->spa_l2cache.sav_config = NULL;
3880		spa_load_l2cache(spa);
3881	}
3882
3883	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3884	    &nvroot) == 0);
3885	if (error == 0)
3886		error = spa_validate_aux(spa, nvroot, -1ULL,
3887		    VDEV_ALLOC_SPARE);
3888	if (error == 0)
3889		error = spa_validate_aux(spa, nvroot, -1ULL,
3890		    VDEV_ALLOC_L2CACHE);
3891	spa_config_exit(spa, SCL_ALL, FTAG);
3892
3893	if (props != NULL)
3894		spa_configfile_set(spa, props, B_FALSE);
3895
3896	if (error != 0 || (props && spa_writeable(spa) &&
3897	    (error = spa_prop_set(spa, props)))) {
3898		spa_unload(spa);
3899		spa_deactivate(spa);
3900		spa_remove(spa);
3901		mutex_exit(&spa_namespace_lock);
3902		return (error);
3903	}
3904
3905	spa_async_resume(spa);
3906
3907	/*
3908	 * Override any spares and level 2 cache devices as specified by
3909	 * the user, as these may have correct device names/devids, etc.
3910	 */
3911	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3912	    &spares, &nspares) == 0) {
3913		if (spa->spa_spares.sav_config)
3914			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3915			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3916		else
3917			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3918			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3919		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3920		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3921		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3922		spa_load_spares(spa);
3923		spa_config_exit(spa, SCL_ALL, FTAG);
3924		spa->spa_spares.sav_sync = B_TRUE;
3925	}
3926	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3927	    &l2cache, &nl2cache) == 0) {
3928		if (spa->spa_l2cache.sav_config)
3929			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3930			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3931		else
3932			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3933			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3934		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3935		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3936		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3937		spa_load_l2cache(spa);
3938		spa_config_exit(spa, SCL_ALL, FTAG);
3939		spa->spa_l2cache.sav_sync = B_TRUE;
3940	}
3941
3942	/*
3943	 * Check for any removed devices.
3944	 */
3945	if (spa->spa_autoreplace) {
3946		spa_aux_check_removed(&spa->spa_spares);
3947		spa_aux_check_removed(&spa->spa_l2cache);
3948	}
3949
3950	if (spa_writeable(spa)) {
3951		/*
3952		 * Update the config cache to include the newly-imported pool.
3953		 */
3954		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3955	}
3956
3957	/*
3958	 * It's possible that the pool was expanded while it was exported.
3959	 * We kick off an async task to handle this for us.
3960	 */
3961	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3962
3963	mutex_exit(&spa_namespace_lock);
3964	spa_history_log_version(spa, "import");
3965
3966	return (0);
3967}
3968
3969nvlist_t *
3970spa_tryimport(nvlist_t *tryconfig)
3971{
3972	nvlist_t *config = NULL;
3973	char *poolname;
3974	spa_t *spa;
3975	uint64_t state;
3976	int error;
3977
3978	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3979		return (NULL);
3980
3981	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3982		return (NULL);
3983
3984	/*
3985	 * Create and initialize the spa structure.
3986	 */
3987	mutex_enter(&spa_namespace_lock);
3988	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3989	spa_activate(spa, FREAD);
3990
3991	/*
3992	 * Pass off the heavy lifting to spa_load().
3993	 * Pass TRUE for mosconfig because the user-supplied config
3994	 * is actually the one to trust when doing an import.
3995	 */
3996	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3997
3998	/*
3999	 * If 'tryconfig' was at least parsable, return the current config.
4000	 */
4001	if (spa->spa_root_vdev != NULL) {
4002		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4003		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4004		    poolname) == 0);
4005		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4006		    state) == 0);
4007		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4008		    spa->spa_uberblock.ub_timestamp) == 0);
4009		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4010		    spa->spa_load_info) == 0);
4011
4012		/*
4013		 * If the bootfs property exists on this pool then we
4014		 * copy it out so that external consumers can tell which
4015		 * pools are bootable.
4016		 */
4017		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4018			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4019
4020			/*
4021			 * We have to play games with the name since the
4022			 * pool was opened as TRYIMPORT_NAME.
4023			 */
4024			if (dsl_dsobj_to_dsname(spa_name(spa),
4025			    spa->spa_bootfs, tmpname) == 0) {
4026				char *cp;
4027				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4028
4029				cp = strchr(tmpname, '/');
4030				if (cp == NULL) {
4031					(void) strlcpy(dsname, tmpname,
4032					    MAXPATHLEN);
4033				} else {
4034					(void) snprintf(dsname, MAXPATHLEN,
4035					    "%s/%s", poolname, ++cp);
4036				}
4037				VERIFY(nvlist_add_string(config,
4038				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4039				kmem_free(dsname, MAXPATHLEN);
4040			}
4041			kmem_free(tmpname, MAXPATHLEN);
4042		}
4043
4044		/*
4045		 * Add the list of hot spares and level 2 cache devices.
4046		 */
4047		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4048		spa_add_spares(spa, config);
4049		spa_add_l2cache(spa, config);
4050		spa_config_exit(spa, SCL_CONFIG, FTAG);
4051	}
4052
4053	spa_unload(spa);
4054	spa_deactivate(spa);
4055	spa_remove(spa);
4056	mutex_exit(&spa_namespace_lock);
4057
4058	return (config);
4059}
4060
4061/*
4062 * Pool export/destroy
4063 *
4064 * The act of destroying or exporting a pool is very simple.  We make sure there
4065 * is no more pending I/O and any references to the pool are gone.  Then, we
4066 * update the pool state and sync all the labels to disk, removing the
4067 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4068 * we don't sync the labels or remove the configuration cache.
4069 */
4070static int
4071spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4072    boolean_t force, boolean_t hardforce)
4073{
4074	spa_t *spa;
4075
4076	if (oldconfig)
4077		*oldconfig = NULL;
4078
4079	if (!(spa_mode_global & FWRITE))
4080		return (SET_ERROR(EROFS));
4081
4082	mutex_enter(&spa_namespace_lock);
4083	if ((spa = spa_lookup(pool)) == NULL) {
4084		mutex_exit(&spa_namespace_lock);
4085		return (SET_ERROR(ENOENT));
4086	}
4087
4088	/*
4089	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4090	 * reacquire the namespace lock, and see if we can export.
4091	 */
4092	spa_open_ref(spa, FTAG);
4093	mutex_exit(&spa_namespace_lock);
4094	spa_async_suspend(spa);
4095	mutex_enter(&spa_namespace_lock);
4096	spa_close(spa, FTAG);
4097
4098	/*
4099	 * The pool will be in core if it's openable,
4100	 * in which case we can modify its state.
4101	 */
4102	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4103		/*
4104		 * Objsets may be open only because they're dirty, so we
4105		 * have to force it to sync before checking spa_refcnt.
4106		 */
4107		txg_wait_synced(spa->spa_dsl_pool, 0);
4108
4109		/*
4110		 * A pool cannot be exported or destroyed if there are active
4111		 * references.  If we are resetting a pool, allow references by
4112		 * fault injection handlers.
4113		 */
4114		if (!spa_refcount_zero(spa) ||
4115		    (spa->spa_inject_ref != 0 &&
4116		    new_state != POOL_STATE_UNINITIALIZED)) {
4117			spa_async_resume(spa);
4118			mutex_exit(&spa_namespace_lock);
4119			return (SET_ERROR(EBUSY));
4120		}
4121
4122		/*
4123		 * A pool cannot be exported if it has an active shared spare.
4124		 * This is to prevent other pools stealing the active spare
4125		 * from an exported pool. At user's own will, such pool can
4126		 * be forcedly exported.
4127		 */
4128		if (!force && new_state == POOL_STATE_EXPORTED &&
4129		    spa_has_active_shared_spare(spa)) {
4130			spa_async_resume(spa);
4131			mutex_exit(&spa_namespace_lock);
4132			return (SET_ERROR(EXDEV));
4133		}
4134
4135		/*
4136		 * We want this to be reflected on every label,
4137		 * so mark them all dirty.  spa_unload() will do the
4138		 * final sync that pushes these changes out.
4139		 */
4140		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4141			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4142			spa->spa_state = new_state;
4143			spa->spa_final_txg = spa_last_synced_txg(spa) +
4144			    TXG_DEFER_SIZE + 1;
4145			vdev_config_dirty(spa->spa_root_vdev);
4146			spa_config_exit(spa, SCL_ALL, FTAG);
4147		}
4148	}
4149
4150	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4151
4152	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4153		spa_unload(spa);
4154		spa_deactivate(spa);
4155	}
4156
4157	if (oldconfig && spa->spa_config)
4158		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4159
4160	if (new_state != POOL_STATE_UNINITIALIZED) {
4161		if (!hardforce)
4162			spa_config_sync(spa, B_TRUE, B_TRUE);
4163		spa_remove(spa);
4164	}
4165	mutex_exit(&spa_namespace_lock);
4166
4167	return (0);
4168}
4169
4170/*
4171 * Destroy a storage pool.
4172 */
4173int
4174spa_destroy(char *pool)
4175{
4176	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4177	    B_FALSE, B_FALSE));
4178}
4179
4180/*
4181 * Export a storage pool.
4182 */
4183int
4184spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4185    boolean_t hardforce)
4186{
4187	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4188	    force, hardforce));
4189}
4190
4191/*
4192 * Similar to spa_export(), this unloads the spa_t without actually removing it
4193 * from the namespace in any way.
4194 */
4195int
4196spa_reset(char *pool)
4197{
4198	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4199	    B_FALSE, B_FALSE));
4200}
4201
4202/*
4203 * ==========================================================================
4204 * Device manipulation
4205 * ==========================================================================
4206 */
4207
4208/*
4209 * Add a device to a storage pool.
4210 */
4211int
4212spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4213{
4214	uint64_t txg, id;
4215	int error;
4216	vdev_t *rvd = spa->spa_root_vdev;
4217	vdev_t *vd, *tvd;
4218	nvlist_t **spares, **l2cache;
4219	uint_t nspares, nl2cache;
4220
4221	ASSERT(spa_writeable(spa));
4222
4223	txg = spa_vdev_enter(spa);
4224
4225	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4226	    VDEV_ALLOC_ADD)) != 0)
4227		return (spa_vdev_exit(spa, NULL, txg, error));
4228
4229	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4230
4231	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4232	    &nspares) != 0)
4233		nspares = 0;
4234
4235	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4236	    &nl2cache) != 0)
4237		nl2cache = 0;
4238
4239	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4240		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4241
4242	if (vd->vdev_children != 0 &&
4243	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4244		return (spa_vdev_exit(spa, vd, txg, error));
4245
4246	/*
4247	 * We must validate the spares and l2cache devices after checking the
4248	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4249	 */
4250	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4251		return (spa_vdev_exit(spa, vd, txg, error));
4252
4253	/*
4254	 * Transfer each new top-level vdev from vd to rvd.
4255	 */
4256	for (int c = 0; c < vd->vdev_children; c++) {
4257
4258		/*
4259		 * Set the vdev id to the first hole, if one exists.
4260		 */
4261		for (id = 0; id < rvd->vdev_children; id++) {
4262			if (rvd->vdev_child[id]->vdev_ishole) {
4263				vdev_free(rvd->vdev_child[id]);
4264				break;
4265			}
4266		}
4267		tvd = vd->vdev_child[c];
4268		vdev_remove_child(vd, tvd);
4269		tvd->vdev_id = id;
4270		vdev_add_child(rvd, tvd);
4271		vdev_config_dirty(tvd);
4272	}
4273
4274	if (nspares != 0) {
4275		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4276		    ZPOOL_CONFIG_SPARES);
4277		spa_load_spares(spa);
4278		spa->spa_spares.sav_sync = B_TRUE;
4279	}
4280
4281	if (nl2cache != 0) {
4282		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4283		    ZPOOL_CONFIG_L2CACHE);
4284		spa_load_l2cache(spa);
4285		spa->spa_l2cache.sav_sync = B_TRUE;
4286	}
4287
4288	/*
4289	 * We have to be careful when adding new vdevs to an existing pool.
4290	 * If other threads start allocating from these vdevs before we
4291	 * sync the config cache, and we lose power, then upon reboot we may
4292	 * fail to open the pool because there are DVAs that the config cache
4293	 * can't translate.  Therefore, we first add the vdevs without
4294	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4295	 * and then let spa_config_update() initialize the new metaslabs.
4296	 *
4297	 * spa_load() checks for added-but-not-initialized vdevs, so that
4298	 * if we lose power at any point in this sequence, the remaining
4299	 * steps will be completed the next time we load the pool.
4300	 */
4301	(void) spa_vdev_exit(spa, vd, txg, 0);
4302
4303	mutex_enter(&spa_namespace_lock);
4304	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4305	mutex_exit(&spa_namespace_lock);
4306
4307	return (0);
4308}
4309
4310/*
4311 * Attach a device to a mirror.  The arguments are the path to any device
4312 * in the mirror, and the nvroot for the new device.  If the path specifies
4313 * a device that is not mirrored, we automatically insert the mirror vdev.
4314 *
4315 * If 'replacing' is specified, the new device is intended to replace the
4316 * existing device; in this case the two devices are made into their own
4317 * mirror using the 'replacing' vdev, which is functionally identical to
4318 * the mirror vdev (it actually reuses all the same ops) but has a few
4319 * extra rules: you can't attach to it after it's been created, and upon
4320 * completion of resilvering, the first disk (the one being replaced)
4321 * is automatically detached.
4322 */
4323int
4324spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4325{
4326	uint64_t txg, dtl_max_txg;
4327	vdev_t *rvd = spa->spa_root_vdev;
4328	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4329	vdev_ops_t *pvops;
4330	char *oldvdpath, *newvdpath;
4331	int newvd_isspare;
4332	int error;
4333
4334	ASSERT(spa_writeable(spa));
4335
4336	txg = spa_vdev_enter(spa);
4337
4338	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4339
4340	if (oldvd == NULL)
4341		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4342
4343	if (!oldvd->vdev_ops->vdev_op_leaf)
4344		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4345
4346	pvd = oldvd->vdev_parent;
4347
4348	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4349	    VDEV_ALLOC_ATTACH)) != 0)
4350		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4351
4352	if (newrootvd->vdev_children != 1)
4353		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4354
4355	newvd = newrootvd->vdev_child[0];
4356
4357	if (!newvd->vdev_ops->vdev_op_leaf)
4358		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4359
4360	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4361		return (spa_vdev_exit(spa, newrootvd, txg, error));
4362
4363	/*
4364	 * Spares can't replace logs
4365	 */
4366	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4367		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4368
4369	if (!replacing) {
4370		/*
4371		 * For attach, the only allowable parent is a mirror or the root
4372		 * vdev.
4373		 */
4374		if (pvd->vdev_ops != &vdev_mirror_ops &&
4375		    pvd->vdev_ops != &vdev_root_ops)
4376			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4377
4378		pvops = &vdev_mirror_ops;
4379	} else {
4380		/*
4381		 * Active hot spares can only be replaced by inactive hot
4382		 * spares.
4383		 */
4384		if (pvd->vdev_ops == &vdev_spare_ops &&
4385		    oldvd->vdev_isspare &&
4386		    !spa_has_spare(spa, newvd->vdev_guid))
4387			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4388
4389		/*
4390		 * If the source is a hot spare, and the parent isn't already a
4391		 * spare, then we want to create a new hot spare.  Otherwise, we
4392		 * want to create a replacing vdev.  The user is not allowed to
4393		 * attach to a spared vdev child unless the 'isspare' state is
4394		 * the same (spare replaces spare, non-spare replaces
4395		 * non-spare).
4396		 */
4397		if (pvd->vdev_ops == &vdev_replacing_ops &&
4398		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4399			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4400		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4401		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4402			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4403		}
4404
4405		if (newvd->vdev_isspare)
4406			pvops = &vdev_spare_ops;
4407		else
4408			pvops = &vdev_replacing_ops;
4409	}
4410
4411	/*
4412	 * Make sure the new device is big enough.
4413	 */
4414	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4415		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4416
4417	/*
4418	 * The new device cannot have a higher alignment requirement
4419	 * than the top-level vdev.
4420	 */
4421	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4422		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4423
4424	/*
4425	 * If this is an in-place replacement, update oldvd's path and devid
4426	 * to make it distinguishable from newvd, and unopenable from now on.
4427	 */
4428	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4429		spa_strfree(oldvd->vdev_path);
4430		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4431		    KM_SLEEP);
4432		(void) sprintf(oldvd->vdev_path, "%s/%s",
4433		    newvd->vdev_path, "old");
4434		if (oldvd->vdev_devid != NULL) {
4435			spa_strfree(oldvd->vdev_devid);
4436			oldvd->vdev_devid = NULL;
4437		}
4438	}
4439
4440	/* mark the device being resilvered */
4441	newvd->vdev_resilvering = B_TRUE;
4442
4443	/*
4444	 * If the parent is not a mirror, or if we're replacing, insert the new
4445	 * mirror/replacing/spare vdev above oldvd.
4446	 */
4447	if (pvd->vdev_ops != pvops)
4448		pvd = vdev_add_parent(oldvd, pvops);
4449
4450	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4451	ASSERT(pvd->vdev_ops == pvops);
4452	ASSERT(oldvd->vdev_parent == pvd);
4453
4454	/*
4455	 * Extract the new device from its root and add it to pvd.
4456	 */
4457	vdev_remove_child(newrootvd, newvd);
4458	newvd->vdev_id = pvd->vdev_children;
4459	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4460	vdev_add_child(pvd, newvd);
4461
4462	tvd = newvd->vdev_top;
4463	ASSERT(pvd->vdev_top == tvd);
4464	ASSERT(tvd->vdev_parent == rvd);
4465
4466	vdev_config_dirty(tvd);
4467
4468	/*
4469	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4470	 * for any dmu_sync-ed blocks.  It will propagate upward when
4471	 * spa_vdev_exit() calls vdev_dtl_reassess().
4472	 */
4473	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4474
4475	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4476	    dtl_max_txg - TXG_INITIAL);
4477
4478	if (newvd->vdev_isspare) {
4479		spa_spare_activate(newvd);
4480		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4481	}
4482
4483	oldvdpath = spa_strdup(oldvd->vdev_path);
4484	newvdpath = spa_strdup(newvd->vdev_path);
4485	newvd_isspare = newvd->vdev_isspare;
4486
4487	/*
4488	 * Mark newvd's DTL dirty in this txg.
4489	 */
4490	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4491
4492	/*
4493	 * Restart the resilver
4494	 */
4495	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4496
4497	/*
4498	 * Commit the config
4499	 */
4500	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4501
4502	spa_history_log_internal(spa, "vdev attach", NULL,
4503	    "%s vdev=%s %s vdev=%s",
4504	    replacing && newvd_isspare ? "spare in" :
4505	    replacing ? "replace" : "attach", newvdpath,
4506	    replacing ? "for" : "to", oldvdpath);
4507
4508	spa_strfree(oldvdpath);
4509	spa_strfree(newvdpath);
4510
4511	if (spa->spa_bootfs)
4512		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4513
4514	return (0);
4515}
4516
4517/*
4518 * Detach a device from a mirror or replacing vdev.
4519 * If 'replace_done' is specified, only detach if the parent
4520 * is a replacing vdev.
4521 */
4522int
4523spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4524{
4525	uint64_t txg;
4526	int error;
4527	vdev_t *rvd = spa->spa_root_vdev;
4528	vdev_t *vd, *pvd, *cvd, *tvd;
4529	boolean_t unspare = B_FALSE;
4530	uint64_t unspare_guid = 0;
4531	char *vdpath;
4532
4533	ASSERT(spa_writeable(spa));
4534
4535	txg = spa_vdev_enter(spa);
4536
4537	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4538
4539	if (vd == NULL)
4540		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4541
4542	if (!vd->vdev_ops->vdev_op_leaf)
4543		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4544
4545	pvd = vd->vdev_parent;
4546
4547	/*
4548	 * If the parent/child relationship is not as expected, don't do it.
4549	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4550	 * vdev that's replacing B with C.  The user's intent in replacing
4551	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4552	 * the replace by detaching C, the expected behavior is to end up
4553	 * M(A,B).  But suppose that right after deciding to detach C,
4554	 * the replacement of B completes.  We would have M(A,C), and then
4555	 * ask to detach C, which would leave us with just A -- not what
4556	 * the user wanted.  To prevent this, we make sure that the
4557	 * parent/child relationship hasn't changed -- in this example,
4558	 * that C's parent is still the replacing vdev R.
4559	 */
4560	if (pvd->vdev_guid != pguid && pguid != 0)
4561		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4562
4563	/*
4564	 * Only 'replacing' or 'spare' vdevs can be replaced.
4565	 */
4566	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4567	    pvd->vdev_ops != &vdev_spare_ops)
4568		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4569
4570	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4571	    spa_version(spa) >= SPA_VERSION_SPARES);
4572
4573	/*
4574	 * Only mirror, replacing, and spare vdevs support detach.
4575	 */
4576	if (pvd->vdev_ops != &vdev_replacing_ops &&
4577	    pvd->vdev_ops != &vdev_mirror_ops &&
4578	    pvd->vdev_ops != &vdev_spare_ops)
4579		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4580
4581	/*
4582	 * If this device has the only valid copy of some data,
4583	 * we cannot safely detach it.
4584	 */
4585	if (vdev_dtl_required(vd))
4586		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4587
4588	ASSERT(pvd->vdev_children >= 2);
4589
4590	/*
4591	 * If we are detaching the second disk from a replacing vdev, then
4592	 * check to see if we changed the original vdev's path to have "/old"
4593	 * at the end in spa_vdev_attach().  If so, undo that change now.
4594	 */
4595	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4596	    vd->vdev_path != NULL) {
4597		size_t len = strlen(vd->vdev_path);
4598
4599		for (int c = 0; c < pvd->vdev_children; c++) {
4600			cvd = pvd->vdev_child[c];
4601
4602			if (cvd == vd || cvd->vdev_path == NULL)
4603				continue;
4604
4605			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4606			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4607				spa_strfree(cvd->vdev_path);
4608				cvd->vdev_path = spa_strdup(vd->vdev_path);
4609				break;
4610			}
4611		}
4612	}
4613
4614	/*
4615	 * If we are detaching the original disk from a spare, then it implies
4616	 * that the spare should become a real disk, and be removed from the
4617	 * active spare list for the pool.
4618	 */
4619	if (pvd->vdev_ops == &vdev_spare_ops &&
4620	    vd->vdev_id == 0 &&
4621	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4622		unspare = B_TRUE;
4623
4624	/*
4625	 * Erase the disk labels so the disk can be used for other things.
4626	 * This must be done after all other error cases are handled,
4627	 * but before we disembowel vd (so we can still do I/O to it).
4628	 * But if we can't do it, don't treat the error as fatal --
4629	 * it may be that the unwritability of the disk is the reason
4630	 * it's being detached!
4631	 */
4632	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4633
4634	/*
4635	 * Remove vd from its parent and compact the parent's children.
4636	 */
4637	vdev_remove_child(pvd, vd);
4638	vdev_compact_children(pvd);
4639
4640	/*
4641	 * Remember one of the remaining children so we can get tvd below.
4642	 */
4643	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4644
4645	/*
4646	 * If we need to remove the remaining child from the list of hot spares,
4647	 * do it now, marking the vdev as no longer a spare in the process.
4648	 * We must do this before vdev_remove_parent(), because that can
4649	 * change the GUID if it creates a new toplevel GUID.  For a similar
4650	 * reason, we must remove the spare now, in the same txg as the detach;
4651	 * otherwise someone could attach a new sibling, change the GUID, and
4652	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4653	 */
4654	if (unspare) {
4655		ASSERT(cvd->vdev_isspare);
4656		spa_spare_remove(cvd);
4657		unspare_guid = cvd->vdev_guid;
4658		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4659		cvd->vdev_unspare = B_TRUE;
4660	}
4661
4662	/*
4663	 * If the parent mirror/replacing vdev only has one child,
4664	 * the parent is no longer needed.  Remove it from the tree.
4665	 */
4666	if (pvd->vdev_children == 1) {
4667		if (pvd->vdev_ops == &vdev_spare_ops)
4668			cvd->vdev_unspare = B_FALSE;
4669		vdev_remove_parent(cvd);
4670		cvd->vdev_resilvering = B_FALSE;
4671	}
4672
4673
4674	/*
4675	 * We don't set tvd until now because the parent we just removed
4676	 * may have been the previous top-level vdev.
4677	 */
4678	tvd = cvd->vdev_top;
4679	ASSERT(tvd->vdev_parent == rvd);
4680
4681	/*
4682	 * Reevaluate the parent vdev state.
4683	 */
4684	vdev_propagate_state(cvd);
4685
4686	/*
4687	 * If the 'autoexpand' property is set on the pool then automatically
4688	 * try to expand the size of the pool. For example if the device we
4689	 * just detached was smaller than the others, it may be possible to
4690	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4691	 * first so that we can obtain the updated sizes of the leaf vdevs.
4692	 */
4693	if (spa->spa_autoexpand) {
4694		vdev_reopen(tvd);
4695		vdev_expand(tvd, txg);
4696	}
4697
4698	vdev_config_dirty(tvd);
4699
4700	/*
4701	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4702	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4703	 * But first make sure we're not on any *other* txg's DTL list, to
4704	 * prevent vd from being accessed after it's freed.
4705	 */
4706	vdpath = spa_strdup(vd->vdev_path);
4707	for (int t = 0; t < TXG_SIZE; t++)
4708		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4709	vd->vdev_detached = B_TRUE;
4710	vdev_dirty(tvd, VDD_DTL, vd, txg);
4711
4712	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4713
4714	/* hang on to the spa before we release the lock */
4715	spa_open_ref(spa, FTAG);
4716
4717	error = spa_vdev_exit(spa, vd, txg, 0);
4718
4719	spa_history_log_internal(spa, "detach", NULL,
4720	    "vdev=%s", vdpath);
4721	spa_strfree(vdpath);
4722
4723	/*
4724	 * If this was the removal of the original device in a hot spare vdev,
4725	 * then we want to go through and remove the device from the hot spare
4726	 * list of every other pool.
4727	 */
4728	if (unspare) {
4729		spa_t *altspa = NULL;
4730
4731		mutex_enter(&spa_namespace_lock);
4732		while ((altspa = spa_next(altspa)) != NULL) {
4733			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4734			    altspa == spa)
4735				continue;
4736
4737			spa_open_ref(altspa, FTAG);
4738			mutex_exit(&spa_namespace_lock);
4739			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4740			mutex_enter(&spa_namespace_lock);
4741			spa_close(altspa, FTAG);
4742		}
4743		mutex_exit(&spa_namespace_lock);
4744
4745		/* search the rest of the vdevs for spares to remove */
4746		spa_vdev_resilver_done(spa);
4747	}
4748
4749	/* all done with the spa; OK to release */
4750	mutex_enter(&spa_namespace_lock);
4751	spa_close(spa, FTAG);
4752	mutex_exit(&spa_namespace_lock);
4753
4754	return (error);
4755}
4756
4757/*
4758 * Split a set of devices from their mirrors, and create a new pool from them.
4759 */
4760int
4761spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4762    nvlist_t *props, boolean_t exp)
4763{
4764	int error = 0;
4765	uint64_t txg, *glist;
4766	spa_t *newspa;
4767	uint_t c, children, lastlog;
4768	nvlist_t **child, *nvl, *tmp;
4769	dmu_tx_t *tx;
4770	char *altroot = NULL;
4771	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4772	boolean_t activate_slog;
4773
4774	ASSERT(spa_writeable(spa));
4775
4776	txg = spa_vdev_enter(spa);
4777
4778	/* clear the log and flush everything up to now */
4779	activate_slog = spa_passivate_log(spa);
4780	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4781	error = spa_offline_log(spa);
4782	txg = spa_vdev_config_enter(spa);
4783
4784	if (activate_slog)
4785		spa_activate_log(spa);
4786
4787	if (error != 0)
4788		return (spa_vdev_exit(spa, NULL, txg, error));
4789
4790	/* check new spa name before going any further */
4791	if (spa_lookup(newname) != NULL)
4792		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4793
4794	/*
4795	 * scan through all the children to ensure they're all mirrors
4796	 */
4797	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4798	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4799	    &children) != 0)
4800		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4801
4802	/* first, check to ensure we've got the right child count */
4803	rvd = spa->spa_root_vdev;
4804	lastlog = 0;
4805	for (c = 0; c < rvd->vdev_children; c++) {
4806		vdev_t *vd = rvd->vdev_child[c];
4807
4808		/* don't count the holes & logs as children */
4809		if (vd->vdev_islog || vd->vdev_ishole) {
4810			if (lastlog == 0)
4811				lastlog = c;
4812			continue;
4813		}
4814
4815		lastlog = 0;
4816	}
4817	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4818		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4819
4820	/* next, ensure no spare or cache devices are part of the split */
4821	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4822	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4823		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4824
4825	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4826	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4827
4828	/* then, loop over each vdev and validate it */
4829	for (c = 0; c < children; c++) {
4830		uint64_t is_hole = 0;
4831
4832		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4833		    &is_hole);
4834
4835		if (is_hole != 0) {
4836			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4837			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4838				continue;
4839			} else {
4840				error = SET_ERROR(EINVAL);
4841				break;
4842			}
4843		}
4844
4845		/* which disk is going to be split? */
4846		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4847		    &glist[c]) != 0) {
4848			error = SET_ERROR(EINVAL);
4849			break;
4850		}
4851
4852		/* look it up in the spa */
4853		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4854		if (vml[c] == NULL) {
4855			error = SET_ERROR(ENODEV);
4856			break;
4857		}
4858
4859		/* make sure there's nothing stopping the split */
4860		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4861		    vml[c]->vdev_islog ||
4862		    vml[c]->vdev_ishole ||
4863		    vml[c]->vdev_isspare ||
4864		    vml[c]->vdev_isl2cache ||
4865		    !vdev_writeable(vml[c]) ||
4866		    vml[c]->vdev_children != 0 ||
4867		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4868		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4869			error = SET_ERROR(EINVAL);
4870			break;
4871		}
4872
4873		if (vdev_dtl_required(vml[c])) {
4874			error = SET_ERROR(EBUSY);
4875			break;
4876		}
4877
4878		/* we need certain info from the top level */
4879		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4880		    vml[c]->vdev_top->vdev_ms_array) == 0);
4881		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4882		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4883		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4884		    vml[c]->vdev_top->vdev_asize) == 0);
4885		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4886		    vml[c]->vdev_top->vdev_ashift) == 0);
4887	}
4888
4889	if (error != 0) {
4890		kmem_free(vml, children * sizeof (vdev_t *));
4891		kmem_free(glist, children * sizeof (uint64_t));
4892		return (spa_vdev_exit(spa, NULL, txg, error));
4893	}
4894
4895	/* stop writers from using the disks */
4896	for (c = 0; c < children; c++) {
4897		if (vml[c] != NULL)
4898			vml[c]->vdev_offline = B_TRUE;
4899	}
4900	vdev_reopen(spa->spa_root_vdev);
4901
4902	/*
4903	 * Temporarily record the splitting vdevs in the spa config.  This
4904	 * will disappear once the config is regenerated.
4905	 */
4906	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4907	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4908	    glist, children) == 0);
4909	kmem_free(glist, children * sizeof (uint64_t));
4910
4911	mutex_enter(&spa->spa_props_lock);
4912	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4913	    nvl) == 0);
4914	mutex_exit(&spa->spa_props_lock);
4915	spa->spa_config_splitting = nvl;
4916	vdev_config_dirty(spa->spa_root_vdev);
4917
4918	/* configure and create the new pool */
4919	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4920	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4921	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4922	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4923	    spa_version(spa)) == 0);
4924	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4925	    spa->spa_config_txg) == 0);
4926	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4927	    spa_generate_guid(NULL)) == 0);
4928	(void) nvlist_lookup_string(props,
4929	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4930
4931	/* add the new pool to the namespace */
4932	newspa = spa_add(newname, config, altroot);
4933	newspa->spa_config_txg = spa->spa_config_txg;
4934	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4935
4936	/* release the spa config lock, retaining the namespace lock */
4937	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4938
4939	if (zio_injection_enabled)
4940		zio_handle_panic_injection(spa, FTAG, 1);
4941
4942	spa_activate(newspa, spa_mode_global);
4943	spa_async_suspend(newspa);
4944
4945	/* create the new pool from the disks of the original pool */
4946	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4947	if (error)
4948		goto out;
4949
4950	/* if that worked, generate a real config for the new pool */
4951	if (newspa->spa_root_vdev != NULL) {
4952		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4953		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4954		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4955		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4956		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4957		    B_TRUE));
4958	}
4959
4960	/* set the props */
4961	if (props != NULL) {
4962		spa_configfile_set(newspa, props, B_FALSE);
4963		error = spa_prop_set(newspa, props);
4964		if (error)
4965			goto out;
4966	}
4967
4968	/* flush everything */
4969	txg = spa_vdev_config_enter(newspa);
4970	vdev_config_dirty(newspa->spa_root_vdev);
4971	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4972
4973	if (zio_injection_enabled)
4974		zio_handle_panic_injection(spa, FTAG, 2);
4975
4976	spa_async_resume(newspa);
4977
4978	/* finally, update the original pool's config */
4979	txg = spa_vdev_config_enter(spa);
4980	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4981	error = dmu_tx_assign(tx, TXG_WAIT);
4982	if (error != 0)
4983		dmu_tx_abort(tx);
4984	for (c = 0; c < children; c++) {
4985		if (vml[c] != NULL) {
4986			vdev_split(vml[c]);
4987			if (error == 0)
4988				spa_history_log_internal(spa, "detach", tx,
4989				    "vdev=%s", vml[c]->vdev_path);
4990			vdev_free(vml[c]);
4991		}
4992	}
4993	vdev_config_dirty(spa->spa_root_vdev);
4994	spa->spa_config_splitting = NULL;
4995	nvlist_free(nvl);
4996	if (error == 0)
4997		dmu_tx_commit(tx);
4998	(void) spa_vdev_exit(spa, NULL, txg, 0);
4999
5000	if (zio_injection_enabled)
5001		zio_handle_panic_injection(spa, FTAG, 3);
5002
5003	/* split is complete; log a history record */
5004	spa_history_log_internal(newspa, "split", NULL,
5005	    "from pool %s", spa_name(spa));
5006
5007	kmem_free(vml, children * sizeof (vdev_t *));
5008
5009	/* if we're not going to mount the filesystems in userland, export */
5010	if (exp)
5011		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5012		    B_FALSE, B_FALSE);
5013
5014	return (error);
5015
5016out:
5017	spa_unload(newspa);
5018	spa_deactivate(newspa);
5019	spa_remove(newspa);
5020
5021	txg = spa_vdev_config_enter(spa);
5022
5023	/* re-online all offlined disks */
5024	for (c = 0; c < children; c++) {
5025		if (vml[c] != NULL)
5026			vml[c]->vdev_offline = B_FALSE;
5027	}
5028	vdev_reopen(spa->spa_root_vdev);
5029
5030	nvlist_free(spa->spa_config_splitting);
5031	spa->spa_config_splitting = NULL;
5032	(void) spa_vdev_exit(spa, NULL, txg, error);
5033
5034	kmem_free(vml, children * sizeof (vdev_t *));
5035	return (error);
5036}
5037
5038static nvlist_t *
5039spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5040{
5041	for (int i = 0; i < count; i++) {
5042		uint64_t guid;
5043
5044		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5045		    &guid) == 0);
5046
5047		if (guid == target_guid)
5048			return (nvpp[i]);
5049	}
5050
5051	return (NULL);
5052}
5053
5054static void
5055spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5056	nvlist_t *dev_to_remove)
5057{
5058	nvlist_t **newdev = NULL;
5059
5060	if (count > 1)
5061		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5062
5063	for (int i = 0, j = 0; i < count; i++) {
5064		if (dev[i] == dev_to_remove)
5065			continue;
5066		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5067	}
5068
5069	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5070	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5071
5072	for (int i = 0; i < count - 1; i++)
5073		nvlist_free(newdev[i]);
5074
5075	if (count > 1)
5076		kmem_free(newdev, (count - 1) * sizeof (void *));
5077}
5078
5079/*
5080 * Evacuate the device.
5081 */
5082static int
5083spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5084{
5085	uint64_t txg;
5086	int error = 0;
5087
5088	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5089	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5090	ASSERT(vd == vd->vdev_top);
5091
5092	/*
5093	 * Evacuate the device.  We don't hold the config lock as writer
5094	 * since we need to do I/O but we do keep the
5095	 * spa_namespace_lock held.  Once this completes the device
5096	 * should no longer have any blocks allocated on it.
5097	 */
5098	if (vd->vdev_islog) {
5099		if (vd->vdev_stat.vs_alloc != 0)
5100			error = spa_offline_log(spa);
5101	} else {
5102		error = SET_ERROR(ENOTSUP);
5103	}
5104
5105	if (error)
5106		return (error);
5107
5108	/*
5109	 * The evacuation succeeded.  Remove any remaining MOS metadata
5110	 * associated with this vdev, and wait for these changes to sync.
5111	 */
5112	ASSERT0(vd->vdev_stat.vs_alloc);
5113	txg = spa_vdev_config_enter(spa);
5114	vd->vdev_removing = B_TRUE;
5115	vdev_dirty(vd, 0, NULL, txg);
5116	vdev_config_dirty(vd);
5117	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5118
5119	return (0);
5120}
5121
5122/*
5123 * Complete the removal by cleaning up the namespace.
5124 */
5125static void
5126spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5127{
5128	vdev_t *rvd = spa->spa_root_vdev;
5129	uint64_t id = vd->vdev_id;
5130	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5131
5132	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5133	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5134	ASSERT(vd == vd->vdev_top);
5135
5136	/*
5137	 * Only remove any devices which are empty.
5138	 */
5139	if (vd->vdev_stat.vs_alloc != 0)
5140		return;
5141
5142	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5143
5144	if (list_link_active(&vd->vdev_state_dirty_node))
5145		vdev_state_clean(vd);
5146	if (list_link_active(&vd->vdev_config_dirty_node))
5147		vdev_config_clean(vd);
5148
5149	vdev_free(vd);
5150
5151	if (last_vdev) {
5152		vdev_compact_children(rvd);
5153	} else {
5154		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5155		vdev_add_child(rvd, vd);
5156	}
5157	vdev_config_dirty(rvd);
5158
5159	/*
5160	 * Reassess the health of our root vdev.
5161	 */
5162	vdev_reopen(rvd);
5163}
5164
5165/*
5166 * Remove a device from the pool -
5167 *
5168 * Removing a device from the vdev namespace requires several steps
5169 * and can take a significant amount of time.  As a result we use
5170 * the spa_vdev_config_[enter/exit] functions which allow us to
5171 * grab and release the spa_config_lock while still holding the namespace
5172 * lock.  During each step the configuration is synced out.
5173 */
5174
5175/*
5176 * Remove a device from the pool.  Currently, this supports removing only hot
5177 * spares, slogs, and level 2 ARC devices.
5178 */
5179int
5180spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5181{
5182	vdev_t *vd;
5183	metaslab_group_t *mg;
5184	nvlist_t **spares, **l2cache, *nv;
5185	uint64_t txg = 0;
5186	uint_t nspares, nl2cache;
5187	int error = 0;
5188	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5189
5190	ASSERT(spa_writeable(spa));
5191
5192	if (!locked)
5193		txg = spa_vdev_enter(spa);
5194
5195	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5196
5197	if (spa->spa_spares.sav_vdevs != NULL &&
5198	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5199	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5200	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5201		/*
5202		 * Only remove the hot spare if it's not currently in use
5203		 * in this pool.
5204		 */
5205		if (vd == NULL || unspare) {
5206			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5207			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5208			spa_load_spares(spa);
5209			spa->spa_spares.sav_sync = B_TRUE;
5210		} else {
5211			error = SET_ERROR(EBUSY);
5212		}
5213	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5214	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5215	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5216	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5217		/*
5218		 * Cache devices can always be removed.
5219		 */
5220		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5221		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5222		spa_load_l2cache(spa);
5223		spa->spa_l2cache.sav_sync = B_TRUE;
5224	} else if (vd != NULL && vd->vdev_islog) {
5225		ASSERT(!locked);
5226		ASSERT(vd == vd->vdev_top);
5227
5228		/*
5229		 * XXX - Once we have bp-rewrite this should
5230		 * become the common case.
5231		 */
5232
5233		mg = vd->vdev_mg;
5234
5235		/*
5236		 * Stop allocating from this vdev.
5237		 */
5238		metaslab_group_passivate(mg);
5239
5240		/*
5241		 * Wait for the youngest allocations and frees to sync,
5242		 * and then wait for the deferral of those frees to finish.
5243		 */
5244		spa_vdev_config_exit(spa, NULL,
5245		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5246
5247		/*
5248		 * Attempt to evacuate the vdev.
5249		 */
5250		error = spa_vdev_remove_evacuate(spa, vd);
5251
5252		txg = spa_vdev_config_enter(spa);
5253
5254		/*
5255		 * If we couldn't evacuate the vdev, unwind.
5256		 */
5257		if (error) {
5258			metaslab_group_activate(mg);
5259			return (spa_vdev_exit(spa, NULL, txg, error));
5260		}
5261
5262		/*
5263		 * Clean up the vdev namespace.
5264		 */
5265		spa_vdev_remove_from_namespace(spa, vd);
5266
5267	} else if (vd != NULL) {
5268		/*
5269		 * Normal vdevs cannot be removed (yet).
5270		 */
5271		error = SET_ERROR(ENOTSUP);
5272	} else {
5273		/*
5274		 * There is no vdev of any kind with the specified guid.
5275		 */
5276		error = SET_ERROR(ENOENT);
5277	}
5278
5279	if (!locked)
5280		return (spa_vdev_exit(spa, NULL, txg, error));
5281
5282	return (error);
5283}
5284
5285/*
5286 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5287 * current spared, so we can detach it.
5288 */
5289static vdev_t *
5290spa_vdev_resilver_done_hunt(vdev_t *vd)
5291{
5292	vdev_t *newvd, *oldvd;
5293
5294	for (int c = 0; c < vd->vdev_children; c++) {
5295		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5296		if (oldvd != NULL)
5297			return (oldvd);
5298	}
5299
5300	/*
5301	 * Check for a completed replacement.  We always consider the first
5302	 * vdev in the list to be the oldest vdev, and the last one to be
5303	 * the newest (see spa_vdev_attach() for how that works).  In
5304	 * the case where the newest vdev is faulted, we will not automatically
5305	 * remove it after a resilver completes.  This is OK as it will require
5306	 * user intervention to determine which disk the admin wishes to keep.
5307	 */
5308	if (vd->vdev_ops == &vdev_replacing_ops) {
5309		ASSERT(vd->vdev_children > 1);
5310
5311		newvd = vd->vdev_child[vd->vdev_children - 1];
5312		oldvd = vd->vdev_child[0];
5313
5314		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5315		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5316		    !vdev_dtl_required(oldvd))
5317			return (oldvd);
5318	}
5319
5320	/*
5321	 * Check for a completed resilver with the 'unspare' flag set.
5322	 */
5323	if (vd->vdev_ops == &vdev_spare_ops) {
5324		vdev_t *first = vd->vdev_child[0];
5325		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5326
5327		if (last->vdev_unspare) {
5328			oldvd = first;
5329			newvd = last;
5330		} else if (first->vdev_unspare) {
5331			oldvd = last;
5332			newvd = first;
5333		} else {
5334			oldvd = NULL;
5335		}
5336
5337		if (oldvd != NULL &&
5338		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5339		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5340		    !vdev_dtl_required(oldvd))
5341			return (oldvd);
5342
5343		/*
5344		 * If there are more than two spares attached to a disk,
5345		 * and those spares are not required, then we want to
5346		 * attempt to free them up now so that they can be used
5347		 * by other pools.  Once we're back down to a single
5348		 * disk+spare, we stop removing them.
5349		 */
5350		if (vd->vdev_children > 2) {
5351			newvd = vd->vdev_child[1];
5352
5353			if (newvd->vdev_isspare && last->vdev_isspare &&
5354			    vdev_dtl_empty(last, DTL_MISSING) &&
5355			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5356			    !vdev_dtl_required(newvd))
5357				return (newvd);
5358		}
5359	}
5360
5361	return (NULL);
5362}
5363
5364static void
5365spa_vdev_resilver_done(spa_t *spa)
5366{
5367	vdev_t *vd, *pvd, *ppvd;
5368	uint64_t guid, sguid, pguid, ppguid;
5369
5370	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5371
5372	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5373		pvd = vd->vdev_parent;
5374		ppvd = pvd->vdev_parent;
5375		guid = vd->vdev_guid;
5376		pguid = pvd->vdev_guid;
5377		ppguid = ppvd->vdev_guid;
5378		sguid = 0;
5379		/*
5380		 * If we have just finished replacing a hot spared device, then
5381		 * we need to detach the parent's first child (the original hot
5382		 * spare) as well.
5383		 */
5384		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5385		    ppvd->vdev_children == 2) {
5386			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5387			sguid = ppvd->vdev_child[1]->vdev_guid;
5388		}
5389		spa_config_exit(spa, SCL_ALL, FTAG);
5390		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5391			return;
5392		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5393			return;
5394		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5395	}
5396
5397	spa_config_exit(spa, SCL_ALL, FTAG);
5398}
5399
5400/*
5401 * Update the stored path or FRU for this vdev.
5402 */
5403int
5404spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5405    boolean_t ispath)
5406{
5407	vdev_t *vd;
5408	boolean_t sync = B_FALSE;
5409
5410	ASSERT(spa_writeable(spa));
5411
5412	spa_vdev_state_enter(spa, SCL_ALL);
5413
5414	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5415		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5416
5417	if (!vd->vdev_ops->vdev_op_leaf)
5418		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5419
5420	if (ispath) {
5421		if (strcmp(value, vd->vdev_path) != 0) {
5422			spa_strfree(vd->vdev_path);
5423			vd->vdev_path = spa_strdup(value);
5424			sync = B_TRUE;
5425		}
5426	} else {
5427		if (vd->vdev_fru == NULL) {
5428			vd->vdev_fru = spa_strdup(value);
5429			sync = B_TRUE;
5430		} else if (strcmp(value, vd->vdev_fru) != 0) {
5431			spa_strfree(vd->vdev_fru);
5432			vd->vdev_fru = spa_strdup(value);
5433			sync = B_TRUE;
5434		}
5435	}
5436
5437	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5438}
5439
5440int
5441spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5442{
5443	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5444}
5445
5446int
5447spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5448{
5449	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5450}
5451
5452/*
5453 * ==========================================================================
5454 * SPA Scanning
5455 * ==========================================================================
5456 */
5457
5458int
5459spa_scan_stop(spa_t *spa)
5460{
5461	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5462	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5463		return (SET_ERROR(EBUSY));
5464	return (dsl_scan_cancel(spa->spa_dsl_pool));
5465}
5466
5467int
5468spa_scan(spa_t *spa, pool_scan_func_t func)
5469{
5470	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5471
5472	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5473		return (SET_ERROR(ENOTSUP));
5474
5475	/*
5476	 * If a resilver was requested, but there is no DTL on a
5477	 * writeable leaf device, we have nothing to do.
5478	 */
5479	if (func == POOL_SCAN_RESILVER &&
5480	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5481		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5482		return (0);
5483	}
5484
5485	return (dsl_scan(spa->spa_dsl_pool, func));
5486}
5487
5488/*
5489 * ==========================================================================
5490 * SPA async task processing
5491 * ==========================================================================
5492 */
5493
5494static void
5495spa_async_remove(spa_t *spa, vdev_t *vd)
5496{
5497	if (vd->vdev_remove_wanted) {
5498		vd->vdev_remove_wanted = B_FALSE;
5499		vd->vdev_delayed_close = B_FALSE;
5500		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5501
5502		/*
5503		 * We want to clear the stats, but we don't want to do a full
5504		 * vdev_clear() as that will cause us to throw away
5505		 * degraded/faulted state as well as attempt to reopen the
5506		 * device, all of which is a waste.
5507		 */
5508		vd->vdev_stat.vs_read_errors = 0;
5509		vd->vdev_stat.vs_write_errors = 0;
5510		vd->vdev_stat.vs_checksum_errors = 0;
5511
5512		vdev_state_dirty(vd->vdev_top);
5513	}
5514
5515	for (int c = 0; c < vd->vdev_children; c++)
5516		spa_async_remove(spa, vd->vdev_child[c]);
5517}
5518
5519static void
5520spa_async_probe(spa_t *spa, vdev_t *vd)
5521{
5522	if (vd->vdev_probe_wanted) {
5523		vd->vdev_probe_wanted = B_FALSE;
5524		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5525	}
5526
5527	for (int c = 0; c < vd->vdev_children; c++)
5528		spa_async_probe(spa, vd->vdev_child[c]);
5529}
5530
5531static void
5532spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5533{
5534	sysevent_id_t eid;
5535	nvlist_t *attr;
5536	char *physpath;
5537
5538	if (!spa->spa_autoexpand)
5539		return;
5540
5541	for (int c = 0; c < vd->vdev_children; c++) {
5542		vdev_t *cvd = vd->vdev_child[c];
5543		spa_async_autoexpand(spa, cvd);
5544	}
5545
5546	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5547		return;
5548
5549	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5550	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5551
5552	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5553	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5554
5555	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5556	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5557
5558	nvlist_free(attr);
5559	kmem_free(physpath, MAXPATHLEN);
5560}
5561
5562static void
5563spa_async_thread(spa_t *spa)
5564{
5565	int tasks;
5566
5567	ASSERT(spa->spa_sync_on);
5568
5569	mutex_enter(&spa->spa_async_lock);
5570	tasks = spa->spa_async_tasks;
5571	spa->spa_async_tasks = 0;
5572	mutex_exit(&spa->spa_async_lock);
5573
5574	/*
5575	 * See if the config needs to be updated.
5576	 */
5577	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5578		uint64_t old_space, new_space;
5579
5580		mutex_enter(&spa_namespace_lock);
5581		old_space = metaslab_class_get_space(spa_normal_class(spa));
5582		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5583		new_space = metaslab_class_get_space(spa_normal_class(spa));
5584		mutex_exit(&spa_namespace_lock);
5585
5586		/*
5587		 * If the pool grew as a result of the config update,
5588		 * then log an internal history event.
5589		 */
5590		if (new_space != old_space) {
5591			spa_history_log_internal(spa, "vdev online", NULL,
5592			    "pool '%s' size: %llu(+%llu)",
5593			    spa_name(spa), new_space, new_space - old_space);
5594		}
5595	}
5596
5597	/*
5598	 * See if any devices need to be marked REMOVED.
5599	 */
5600	if (tasks & SPA_ASYNC_REMOVE) {
5601		spa_vdev_state_enter(spa, SCL_NONE);
5602		spa_async_remove(spa, spa->spa_root_vdev);
5603		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5604			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5605		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5606			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5607		(void) spa_vdev_state_exit(spa, NULL, 0);
5608	}
5609
5610	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5611		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5612		spa_async_autoexpand(spa, spa->spa_root_vdev);
5613		spa_config_exit(spa, SCL_CONFIG, FTAG);
5614	}
5615
5616	/*
5617	 * See if any devices need to be probed.
5618	 */
5619	if (tasks & SPA_ASYNC_PROBE) {
5620		spa_vdev_state_enter(spa, SCL_NONE);
5621		spa_async_probe(spa, spa->spa_root_vdev);
5622		(void) spa_vdev_state_exit(spa, NULL, 0);
5623	}
5624
5625	/*
5626	 * If any devices are done replacing, detach them.
5627	 */
5628	if (tasks & SPA_ASYNC_RESILVER_DONE)
5629		spa_vdev_resilver_done(spa);
5630
5631	/*
5632	 * Kick off a resilver.
5633	 */
5634	if (tasks & SPA_ASYNC_RESILVER)
5635		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5636
5637	/*
5638	 * Let the world know that we're done.
5639	 */
5640	mutex_enter(&spa->spa_async_lock);
5641	spa->spa_async_thread = NULL;
5642	cv_broadcast(&spa->spa_async_cv);
5643	mutex_exit(&spa->spa_async_lock);
5644	thread_exit();
5645}
5646
5647void
5648spa_async_suspend(spa_t *spa)
5649{
5650	mutex_enter(&spa->spa_async_lock);
5651	spa->spa_async_suspended++;
5652	while (spa->spa_async_thread != NULL)
5653		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5654	mutex_exit(&spa->spa_async_lock);
5655}
5656
5657void
5658spa_async_resume(spa_t *spa)
5659{
5660	mutex_enter(&spa->spa_async_lock);
5661	ASSERT(spa->spa_async_suspended != 0);
5662	spa->spa_async_suspended--;
5663	mutex_exit(&spa->spa_async_lock);
5664}
5665
5666static void
5667spa_async_dispatch(spa_t *spa)
5668{
5669	mutex_enter(&spa->spa_async_lock);
5670	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5671	    spa->spa_async_thread == NULL &&
5672	    rootdir != NULL && !vn_is_readonly(rootdir))
5673		spa->spa_async_thread = thread_create(NULL, 0,
5674		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5675	mutex_exit(&spa->spa_async_lock);
5676}
5677
5678void
5679spa_async_request(spa_t *spa, int task)
5680{
5681	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5682	mutex_enter(&spa->spa_async_lock);
5683	spa->spa_async_tasks |= task;
5684	mutex_exit(&spa->spa_async_lock);
5685}
5686
5687/*
5688 * ==========================================================================
5689 * SPA syncing routines
5690 * ==========================================================================
5691 */
5692
5693static int
5694bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5695{
5696	bpobj_t *bpo = arg;
5697	bpobj_enqueue(bpo, bp, tx);
5698	return (0);
5699}
5700
5701static int
5702spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5703{
5704	zio_t *zio = arg;
5705
5706	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5707	    zio->io_flags));
5708	return (0);
5709}
5710
5711static void
5712spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5713{
5714	char *packed = NULL;
5715	size_t bufsize;
5716	size_t nvsize = 0;
5717	dmu_buf_t *db;
5718
5719	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5720
5721	/*
5722	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5723	 * information.  This avoids the dbuf_will_dirty() path and
5724	 * saves us a pre-read to get data we don't actually care about.
5725	 */
5726	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5727	packed = kmem_alloc(bufsize, KM_SLEEP);
5728
5729	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5730	    KM_SLEEP) == 0);
5731	bzero(packed + nvsize, bufsize - nvsize);
5732
5733	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5734
5735	kmem_free(packed, bufsize);
5736
5737	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5738	dmu_buf_will_dirty(db, tx);
5739	*(uint64_t *)db->db_data = nvsize;
5740	dmu_buf_rele(db, FTAG);
5741}
5742
5743static void
5744spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5745    const char *config, const char *entry)
5746{
5747	nvlist_t *nvroot;
5748	nvlist_t **list;
5749	int i;
5750
5751	if (!sav->sav_sync)
5752		return;
5753
5754	/*
5755	 * Update the MOS nvlist describing the list of available devices.
5756	 * spa_validate_aux() will have already made sure this nvlist is
5757	 * valid and the vdevs are labeled appropriately.
5758	 */
5759	if (sav->sav_object == 0) {
5760		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5761		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5762		    sizeof (uint64_t), tx);
5763		VERIFY(zap_update(spa->spa_meta_objset,
5764		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5765		    &sav->sav_object, tx) == 0);
5766	}
5767
5768	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5769	if (sav->sav_count == 0) {
5770		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5771	} else {
5772		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5773		for (i = 0; i < sav->sav_count; i++)
5774			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5775			    B_FALSE, VDEV_CONFIG_L2CACHE);
5776		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5777		    sav->sav_count) == 0);
5778		for (i = 0; i < sav->sav_count; i++)
5779			nvlist_free(list[i]);
5780		kmem_free(list, sav->sav_count * sizeof (void *));
5781	}
5782
5783	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5784	nvlist_free(nvroot);
5785
5786	sav->sav_sync = B_FALSE;
5787}
5788
5789static void
5790spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5791{
5792	nvlist_t *config;
5793
5794	if (list_is_empty(&spa->spa_config_dirty_list))
5795		return;
5796
5797	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5798
5799	config = spa_config_generate(spa, spa->spa_root_vdev,
5800	    dmu_tx_get_txg(tx), B_FALSE);
5801
5802	/*
5803	 * If we're upgrading the spa version then make sure that
5804	 * the config object gets updated with the correct version.
5805	 */
5806	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5807		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5808		    spa->spa_uberblock.ub_version);
5809
5810	spa_config_exit(spa, SCL_STATE, FTAG);
5811
5812	if (spa->spa_config_syncing)
5813		nvlist_free(spa->spa_config_syncing);
5814	spa->spa_config_syncing = config;
5815
5816	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5817}
5818
5819static void
5820spa_sync_version(void *arg, dmu_tx_t *tx)
5821{
5822	uint64_t *versionp = arg;
5823	uint64_t version = *versionp;
5824	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5825
5826	/*
5827	 * Setting the version is special cased when first creating the pool.
5828	 */
5829	ASSERT(tx->tx_txg != TXG_INITIAL);
5830
5831	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5832	ASSERT(version >= spa_version(spa));
5833
5834	spa->spa_uberblock.ub_version = version;
5835	vdev_config_dirty(spa->spa_root_vdev);
5836	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5837}
5838
5839/*
5840 * Set zpool properties.
5841 */
5842static void
5843spa_sync_props(void *arg, dmu_tx_t *tx)
5844{
5845	nvlist_t *nvp = arg;
5846	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5847	objset_t *mos = spa->spa_meta_objset;
5848	nvpair_t *elem = NULL;
5849
5850	mutex_enter(&spa->spa_props_lock);
5851
5852	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5853		uint64_t intval;
5854		char *strval, *fname;
5855		zpool_prop_t prop;
5856		const char *propname;
5857		zprop_type_t proptype;
5858		zfeature_info_t *feature;
5859
5860		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5861		case ZPROP_INVAL:
5862			/*
5863			 * We checked this earlier in spa_prop_validate().
5864			 */
5865			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5866
5867			fname = strchr(nvpair_name(elem), '@') + 1;
5868			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5869
5870			spa_feature_enable(spa, feature, tx);
5871			spa_history_log_internal(spa, "set", tx,
5872			    "%s=enabled", nvpair_name(elem));
5873			break;
5874
5875		case ZPOOL_PROP_VERSION:
5876			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5877			/*
5878			 * The version is synced seperatly before other
5879			 * properties and should be correct by now.
5880			 */
5881			ASSERT3U(spa_version(spa), >=, intval);
5882			break;
5883
5884		case ZPOOL_PROP_ALTROOT:
5885			/*
5886			 * 'altroot' is a non-persistent property. It should
5887			 * have been set temporarily at creation or import time.
5888			 */
5889			ASSERT(spa->spa_root != NULL);
5890			break;
5891
5892		case ZPOOL_PROP_READONLY:
5893		case ZPOOL_PROP_CACHEFILE:
5894			/*
5895			 * 'readonly' and 'cachefile' are also non-persisitent
5896			 * properties.
5897			 */
5898			break;
5899		case ZPOOL_PROP_COMMENT:
5900			VERIFY(nvpair_value_string(elem, &strval) == 0);
5901			if (spa->spa_comment != NULL)
5902				spa_strfree(spa->spa_comment);
5903			spa->spa_comment = spa_strdup(strval);
5904			/*
5905			 * We need to dirty the configuration on all the vdevs
5906			 * so that their labels get updated.  It's unnecessary
5907			 * to do this for pool creation since the vdev's
5908			 * configuratoin has already been dirtied.
5909			 */
5910			if (tx->tx_txg != TXG_INITIAL)
5911				vdev_config_dirty(spa->spa_root_vdev);
5912			spa_history_log_internal(spa, "set", tx,
5913			    "%s=%s", nvpair_name(elem), strval);
5914			break;
5915		default:
5916			/*
5917			 * Set pool property values in the poolprops mos object.
5918			 */
5919			if (spa->spa_pool_props_object == 0) {
5920				spa->spa_pool_props_object =
5921				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5922				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5923				    tx);
5924			}
5925
5926			/* normalize the property name */
5927			propname = zpool_prop_to_name(prop);
5928			proptype = zpool_prop_get_type(prop);
5929
5930			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5931				ASSERT(proptype == PROP_TYPE_STRING);
5932				VERIFY(nvpair_value_string(elem, &strval) == 0);
5933				VERIFY(zap_update(mos,
5934				    spa->spa_pool_props_object, propname,
5935				    1, strlen(strval) + 1, strval, tx) == 0);
5936				spa_history_log_internal(spa, "set", tx,
5937				    "%s=%s", nvpair_name(elem), strval);
5938			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5939				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5940
5941				if (proptype == PROP_TYPE_INDEX) {
5942					const char *unused;
5943					VERIFY(zpool_prop_index_to_string(
5944					    prop, intval, &unused) == 0);
5945				}
5946				VERIFY(zap_update(mos,
5947				    spa->spa_pool_props_object, propname,
5948				    8, 1, &intval, tx) == 0);
5949				spa_history_log_internal(spa, "set", tx,
5950				    "%s=%lld", nvpair_name(elem), intval);
5951			} else {
5952				ASSERT(0); /* not allowed */
5953			}
5954
5955			switch (prop) {
5956			case ZPOOL_PROP_DELEGATION:
5957				spa->spa_delegation = intval;
5958				break;
5959			case ZPOOL_PROP_BOOTFS:
5960				spa->spa_bootfs = intval;
5961				break;
5962			case ZPOOL_PROP_FAILUREMODE:
5963				spa->spa_failmode = intval;
5964				break;
5965			case ZPOOL_PROP_AUTOEXPAND:
5966				spa->spa_autoexpand = intval;
5967				if (tx->tx_txg != TXG_INITIAL)
5968					spa_async_request(spa,
5969					    SPA_ASYNC_AUTOEXPAND);
5970				break;
5971			case ZPOOL_PROP_DEDUPDITTO:
5972				spa->spa_dedup_ditto = intval;
5973				break;
5974			default:
5975				break;
5976			}
5977		}
5978
5979	}
5980
5981	mutex_exit(&spa->spa_props_lock);
5982}
5983
5984/*
5985 * Perform one-time upgrade on-disk changes.  spa_version() does not
5986 * reflect the new version this txg, so there must be no changes this
5987 * txg to anything that the upgrade code depends on after it executes.
5988 * Therefore this must be called after dsl_pool_sync() does the sync
5989 * tasks.
5990 */
5991static void
5992spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5993{
5994	dsl_pool_t *dp = spa->spa_dsl_pool;
5995
5996	ASSERT(spa->spa_sync_pass == 1);
5997
5998	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
5999
6000	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6001	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6002		dsl_pool_create_origin(dp, tx);
6003
6004		/* Keeping the origin open increases spa_minref */
6005		spa->spa_minref += 3;
6006	}
6007
6008	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6009	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6010		dsl_pool_upgrade_clones(dp, tx);
6011	}
6012
6013	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6014	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6015		dsl_pool_upgrade_dir_clones(dp, tx);
6016
6017		/* Keeping the freedir open increases spa_minref */
6018		spa->spa_minref += 3;
6019	}
6020
6021	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6022	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6023		spa_feature_create_zap_objects(spa, tx);
6024	}
6025	rrw_exit(&dp->dp_config_rwlock, FTAG);
6026}
6027
6028/*
6029 * Sync the specified transaction group.  New blocks may be dirtied as
6030 * part of the process, so we iterate until it converges.
6031 */
6032void
6033spa_sync(spa_t *spa, uint64_t txg)
6034{
6035	dsl_pool_t *dp = spa->spa_dsl_pool;
6036	objset_t *mos = spa->spa_meta_objset;
6037	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6038	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6039	vdev_t *rvd = spa->spa_root_vdev;
6040	vdev_t *vd;
6041	dmu_tx_t *tx;
6042	int error;
6043
6044	VERIFY(spa_writeable(spa));
6045
6046	/*
6047	 * Lock out configuration changes.
6048	 */
6049	spa_config_enter(spa, SCL_CONFIG,